@booklet {sharma_robust_2023, title = {Robust Decision-Focused Learning for Reward Transfer}, year = {2023}, note = {arXiv:2304.03365 [cs]}, publisher = {arXiv}, abstract = {Decision-focused (DF) model-based reinforcement learning has recently been introduced as a powerful algorithm which can focus on learning the MDP dynamics which are most relevant for obtaining high rewards. While this approach increases the performance of agents by focusing the learning towards optimizing for the reward directly, it does so by learning less accurate dynamics (from a MLE standpoint), and may thus be brittle to changes in the reward function. In this work, we develop the robust decision-focused (RDF) algorithm which leverages the non-identifiability of DF solutions to learn models which maximize expected returns while simultaneously learning models which are robust to changes in the reward function. We demonstrate on a variety of toy example and healthcare simulators that RDF significantly increases the robustness of DF to changes in the reward function, without decreasing the overall return the agent obtains.}, keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning}, doi = {10.48550/arXiv.2304.03365}, url = {http://arxiv.org/abs/2304.03365}, author = {Abhishek Sharma and Parbhoo, Sonali and Omer Gottesman and Finale Doshi-Velez} } @conference {fu_performance_2023, title = {Performance Bounds for Model and Policy Transfer in Hidden-parameter MDPs}, booktitle = {The Eleventh International Conference on Learning Representations}, year = {2023}, abstract = {In the Hidden-Parameter MDP (HiP-MDP) framework, a family of reinforcement learning tasks is generated by varying hidden parameters specifying the dynamics and reward function for each individual task. HiP-MDP is a natural model for families of tasks in which meta- and lifelong-reinforcement learning approaches can succeed. Given a learned context encoder that infers the hidden parameters from previous experience, most existing algorithms fall into two categories: \$\textbackslashtextit\model transfer\\$ and \$\textbackslashtextit\policy transfer\\$, depending on which function the hidden parameters are used to parameterize. We characterize the robustness of model and policy transfer algorithms with respect to hidden parameter estimation error. We first show that the value function of HiP-MDPs is Lipschitz continuous under certain conditions. We then derive regret bounds for both settings through the lens of Lipschitz continuity. Finally, we empirically corroborate our theoretical analysis by experimentally varying the hyper-parameters governing the Lipschitz constants of two continuous control problems; the resulting performance is consistent with our predictions.}, url = {https://openreview.net/forum?id=20gBzEzgtiI}, author = {Fu, Haotian and Jiayu Yao and Omer Gottesman and Finale Doshi-Velez and George Konidaris} } @article {sharma_robust_2023, title = {Robust Decision-Focused Learning for Reward Transfer}, journal = {Preprint}, year = {2023}, note = {arXiv:2304.03365 [cs]}, publisher = {arXiv}, abstract = {Decision-focused (DF) model-based reinforcement learning has recently been introduced as a powerful algorithm which can focus on learning the MDP dynamics which are most relevant for obtaining high rewards. While this approach increases the performance of agents by focusing the learning towards optimizing for the reward directly, it does so by learning less accurate dynamics (from a MLE standpoint), and may thus be brittle to changes in the reward function. In this work, we develop the robust decision-focused (RDF) algorithm which leverages the non-identifiability of DF solutions to learn models which maximize expected returns while simultaneously learning models which are robust to changes in the reward function. We demonstrate on a variety of toy example and healthcare simulators that RDF significantly increases the robustness of DF to changes in the reward function, without decreasing the overall return the agent obtains.}, keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning}, doi = {10.48550/arXiv.2304.03365}, url = {http://arxiv.org/abs/2304.03365}, author = {Abhishek Sharma and Parbhoo, Sonali and Omer Gottesman and Finale Doshi-Velez} } @article {sharma_travel-time_2023, title = {Travel-time prediction using neural-network-based mixture models}, journal = {The 14th International Conference on Ambient Systems, Networks and Technologies Networks (ANT 2022) and The 6th International Conference on Emerging Data and Industry 4.0 (EDI40)}, volume = {220}, year = {2023}, pages = {1033{\textendash}1038}, abstract = {Accurate estimation of travel times is an important step in smart transportation and smart building systems. Poor estimation of travel times results in both frustrated users and wasted resources. Current methods that estimate travel times usually only return point estimates, losing important distributional information necessary for accurate decision-making. We propose using neural network-based mixture distributions to predict a user{\textquoteright}s travel times given their origin and destination coordinates. We show that our method correctly estimates the travel time distribution, maximizes utility in a downstream elevator scheduling task, and is easy to retrain{\textemdash}making it a versatile and an inexpensive-to-maintain module when deployed in smart crowd management systems.}, keywords = {mixture modeling, Neural networks, travel-time predictions}, issn = {1877-0509}, doi = {10.1016/j.procs.2023.03.144}, url = {https://www.sciencedirect.com/science/article/pii/S1877050923006798}, author = {Abhishek Sharma and Zhang, Jing and Daniel Nikovski and Finale Doshi-Velez} } @article {noauthor_addressing_nodate, title = {Addressing Leakage in Concept Bottleneck Models}, journal = {Preprint}, year = {2022}, month = {2022}, url = {https://openreview.net/forum?id=tglniD_fn9}, author = {Havasi, Marton and Parbhoo, Sonali and Finale Doshi-Velez} } @article {zhang_bayesian_2022, title = {A Bayesian Approach to Learning Bandit Structure in Markov Decision Processes}, journal = {Preprint}, year = {2022}, note = {arXiv:2208.00250 [cs]}, month = {jul}, publisher = {arXiv}, abstract = {In the reinforcement learning literature, there are many algorithms developed for either Contextual Bandit (CB) or Markov Decision Processes (MDP) environments. However, when deploying reinforcement learning algorithms in the real world, even with domain expertise, it is often difficult to know whether it is appropriate to treat a sequential decision making problem as a CB or an MDP. In other words, do actions affect future states, or only the immediate rewards? Making the wrong assumption regarding the nature of the environment can lead to inefficient learning, or even prevent the algorithm from ever learning an optimal policy, even with infinite data. In this work we develop an online algorithm that uses a Bayesian hypothesis testing approach to learn the nature of the environment. Our algorithm allows practitioners to incorporate prior knowledge about whether the environment is that of a CB or an MDP, and effectively interpolate between classical CB and MDP-based algorithms to mitigate against the effects of misspecifying the environment. We perform simulations and demonstrate that in CB settings our algorithm achieves lower regret than MDP-based algorithms, while in non-bandit MDP settings our algorithm is able to learn the optimal policy, often achieving comparable regret to MDP-based algorithms.}, doi = {10.48550/arXiv.2208.00250}, url = {http://arxiv.org/abs/2208.00250}, author = {Zhang, Kelly W. and Omer Gottesman and Finale Doshi-Velez} } @article {liao_connecting_2022, title = {Connecting Algorithmic Research and Usage Contexts: A Perspective of Contextualized Evaluation for Explainable AI}, journal = {Proceedings of the AAAI Conference on Human Computation and Crowdsourcing}, volume = {10}, year = {2022}, month = {oct}, pages = {147{\textendash}159}, abstract = {Recent years have seen a surge of interest in the field of explainable AI (XAI), with a plethora of algorithms proposed in the literature. However, a lack of consensus on how to evaluate XAI hinders the advancement of the field. We highlight that XAI is not a monolithic set of technologies{\textendash}-researchers and practitioners have begun to leverage XAI algorithms to build XAI systems that serve different usage contexts, such as model debugging and decision-support. Algorithmic research of XAI, however, often does not account for these diverse downstream usage contexts, resulting in limited effectiveness or even unintended consequences for actual users, as well as difficulties for practitioners to make technical choices. We argue that one way to close the gap is to develop evaluation methods that account for different user requirements in these usage contexts. Towards this goal, we introduce a perspective of contextualized XAI evaluation by considering the relative importance of XAI evaluation criteria for prototypical usage contexts of XAI. To explore the context dependency of XAI evaluation criteria, we conduct two survey studies, one with XAI topical experts and another with crowd workers. Our results urge for responsible AI research with usage-informed evaluation practices, and provide a nuanced understanding of user requirements for XAI in different usage contexts.}, issn = {2769-1349}, doi = {10.1609/hcomp.v10i1.21995}, url = {https://ojs.aaai.org/index.php/HCOMP/article/view/21995}, author = {Liao, Q. Vera and Zhang, Yunfeng and Luss, Ronny and Finale Doshi-Velez and Dhurandhar, Amit} } @article {trella_designing_2022, title = {Designing Reinforcement Learning Algorithms for Digital Interventions: Pre-Implementation Guidelines}, journal = {Algorithms}, volume = {15}, number = {8}, year = {2022}, note = {Number: 8 Publisher: Multidisciplinary Digital Publishing Institute}, pages = {255}, abstract = {Online reinforcement learning (RL) algorithms are increasingly used to personalize digital interventions in the fields of mobile health and online education. Common challenges in designing and testing an RL algorithm in these settings include ensuring the RL algorithm can learn and run stably under real-time constraints, and accounting for the complexity of the environment, e.g., a lack of accurate mechanistic models for the user dynamics. To guide how one can tackle these challenges, we extend the PCS (predictability, computability, stability) framework, a data science framework that incorporates best practices from machine learning and statistics in supervised learning to the design of RL algorithms for the digital interventions setting. Furthermore, we provide guidelines on how to design simulation environments, a crucial tool for evaluating RL candidate algorithms using the PCS framework. We show how we used the PCS framework to design an RL algorithm for Oralytics, a mobile health study aiming to improve users{\textquoteright} tooth-brushing behaviors through the personalized delivery of intervention messages. Oralytics will go into the field in late 2022.}, issn = {1999-4893}, doi = {10.3390/a15080255}, url = {https://www.mdpi.com/1999-4893/15/8/255}, author = {Anna L. Trella and Zhang, Kelly W. and Nahum-Shani, Inbal and Shetty, Vivek and Finale Doshi-Velez and Murphy, Susan A.} } @article {lage_clinicians_2022, title = {Do clinicians follow heuristics in prescribing antidepressants?}, journal = {Journal of Affective Disorders}, volume = {311}, year = {2022}, month = {aug}, pages = {110{\textendash}114}, abstract = {Background While clinicians commonly learn heuristics to guide antidepressant treatment selection, surveys suggest real-world prescribing practices vary widely. We aimed to determine the extent to which antidepressant prescriptions were consistent with commonly-advocated heuristics for treatment selection. Methods This retrospective longitudinal cohort study examined electronic health records from psychiatry and non-psychiatry practice networks affiliated with two large academic medical centers between March 2008 and December 2017. Patients included 45,955 individuals with a major depressive disorder or depressive disorder not otherwise specified diagnosis who were prescribed at least one of 11 common antidepressant medications. Specific clinical features that may impact prescribing choices were extracted from coded data, and analyzed for association with index prescription in logistic regression models adjusted for sociodemographic variables and provider type. Results Multiple clinical features yielded 10\% or greater change in odds of prescribing, including overweight and underweight status and sexual dysfunction. These heuristics were generally applied similarly across hospital systems and psychiatrist and non-psychiatrist providers. Limitations These analyses rely on coded clinical data, which is likely to substantially underestimate prevalence of particular clinical features. Additionally, numerous other features that may impact prescribing choices are not able to be modeled. Conclusion Our results confirm the hypothesis that clinicians apply heuristics on the basis of clinical features to guide antidepressant prescribing, although the magnitude of these effects is modest, suggesting other patient- or clinician-level factors have larger effects. Funding This work was funded by NSF GRFP (grant no. DGE1745303), Harvard SEAS, the Center for Research on Computation and Society at Harvard, the Harvard Data Science Initiative, and a grant from the National Institute of Mental Health (grant no. 1R01MH106577).}, issn = {0165-0327}, doi = {10.1016/j.jad.2022.04.128}, url = {https://www.sciencedirect.com/science/article/pii/S0165032722004724}, author = {Isaac Lage and Melanie F. Pradier and Thomas H. McCoy and Roy H. Perlis and Finale Doshi-Velez} } @article {lage_efficiently_2022, title = {Efficiently identifying individuals at high risk for treatment resistance in major depressive disorder using electronic health records}, journal = {Journal of Affective Disorders}, volume = {306}, year = {2022}, pages = {254{\textendash}259}, abstract = {Background With the emergence of evidence-based treatments for treatment-resistant depression, strategies to identify individuals at greater risk for treatment resistance early in the course of illness could have clinical utility. We sought to develop and validate a model to predict treatment resistance in major depressive disorder using coded clinical data from the electronic health record. Methods We identified individuals from a large health system with a diagnosis of major depressive disorder receiving an index antidepressant prescription, and used a tree-based machine learning classifier to build a risk stratification model to identify those likely to experience treatment resistance. The resulting model was validated in a second health system. Results In the second health system, the extra trees model yielded an AUC of 0.652 (95\% CI: 0.623{\textendash}0.682); with sensitivity constrained at 0.80, specificity was 0.358 (95\% CI: 0.300{\textendash}0.413). Lift in the top quintile was 1.99 (95\% CI: 1.76{\textendash}2.22). Including additional data for the 4 weeks following treatment initiation did not meaningfully improve model performance. Limitations The extent to which these models generalize across additional health systems will require further investigation. Conclusion Electronic health records facilitated stratification of risk for treatment-resistant depression and demonstrated generalizability to a second health system. Efforts to improve upon such models using additional measures, and to understand their performance in real-world clinical settings, are warranted.}, keywords = {CV - Journals}, issn = {0165-0327}, doi = {10.1016/j.jad.2022.02.046}, url = {https://www.sciencedirect.com/science/article/pii/S0165032722001951}, author = {Isaac Lage and McCoy Jr, Thomas H. and Roy H. Perlis and Finale Doshi-Velez} } @article {zeng_soft_nodate, title = {From Soft Trees to Hard Trees: Gains and Losses}, journal = {Preprint}, year = {2022}, abstract = {Trees are widely used as interpretable models. However, when they are greedily trained they can yield suboptimal predictive performance. Training soft trees, with probabilistic splits rather than deterministic ones, provides a way to supposedly globally optimize tree models. For interpretability purposes, a hard tree can be obtained from a soft tree by binarizing the probabilistic splits, called hardening. Unfortunately, the good performance of the soft model is often lost after hardening. We systematically study two factors contributing to the performance drop: first, the loss surface of the soft tree loss has many local optima (and thus the logic for using the soft tree loss becomes less clear), and second, the relative values of the soft tree loss do not correspond to relative values of the hard tree loss. We also demonstrate that simple mitigation methods in literature do not fully mitigate the performance drop.}, author = {Zeng, Xin and Jiayu Yao and Finale Doshi-Velez and Weiwei Pan} } @article {littman_gathering_2022, title = {Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report}, journal = {Preprint}, year = {2022}, note = {arXiv:2210.15767 [cs]}, publisher = {arXiv}, abstract = {In September 2021, the "One Hundred Year Study on Artificial Intelligence" project (AI100) issued the second report of its planned long-term periodic assessment of artificial intelligence (AI) and its impact on society. It was written by a panel of 17 study authors, each of whom is deeply rooted in AI research, chaired by Michael Littman of Brown University. The report, entitled "Gathering Strength, Gathering Storms," answers a set of 14 questions probing critical areas of AI development addressing the major risks and dangers of AI, its effects on society, its public perception and the future of the field. The report concludes that AI has made a major leap from the lab to people{\textquoteright}s lives in recent years, which increases the urgency to understand its potential negative effects. The questions were developed by the AI100 Standing Committee, chaired by Peter Stone of the University of Texas at Austin, consisting of a group of AI leaders with expertise in computer science, sociology, ethics, economics, and other disciplines.}, doi = {10.48550/arXiv.2210.15767}, url = {http://arxiv.org/abs/2210.15767}, author = {Littman, Michael L. and Ajunwa, Ifeoma and Berger, Guy and Craig Boutilier and Currie, Morgan and Finale Doshi-Velez and Hadfield, Gillian and Horowitz, Michael C. and Isbell, Charles and Kitano, Hiroaki and Levy, Karen and Lyons, Terah and Mitchell, Melanie and Shah, Julie and Sloman, Steven and Vallor, Shannon and Toby Walsh} } @article {parbhoo_generalizing_2022, title = {Generalizing Off-Policy Evaluation From a Causal Perspective For Sequential Decision-Making}, journal = {Preprint}, year = {2022}, note = {arXiv:2201.08262 [cs, stat]}, publisher = {arXiv}, abstract = {Assessing the effects of a policy based on observational data from a different policy is a common problem across several high-stake decision-making domains, and several off-policy evaluation (OPE) techniques have been proposed. However, these methods largely formulate OPE as a problem disassociated from the process used to generate the data (i.e. structural assumptions in the form of a causal graph). We argue that explicitly highlighting this association has important implications on our understanding of the fundamental limits of OPE. First, this implies that current formulation of OPE corresponds to a narrow set of tasks, i.e. a specific causal estimand which is focused on prospective evaluation of policies over populations or sub-populations. Second, we demonstrate how this association motivates natural desiderata to consider a general set of causal estimands, particularly extending the role of OPE for counterfactual off-policy evaluation at the level of individuals of the population. A precise description of the causal estimand highlights which OPE estimands are identifiable from observational data under the stated generative assumptions. For those OPE estimands that are not identifiable, the causal perspective further highlights where more experimental data is necessary, and highlights situations where human expertise can aid identification and estimation. Furthermore, many formalisms of OPE overlook the role of uncertainty entirely in the estimation process.We demonstrate how specifically characterising the causal estimand highlights the different sources of uncertainty and when human expertise can naturally manage this uncertainty. We discuss each of these aspects as actionable desiderata for future OPE research at scale and in-line with practical utility.}, doi = {10.48550/arXiv.2201.08262}, url = {http://arxiv.org/abs/2201.08262}, author = {Parbhoo, Sonali and Joshi, Shalmali and Finale Doshi-Velez} } @article {keramati_identification_nodate, title = {Identification of Subgroups With Similar Benefits in Off-Policy Policy Evaluation}, journal = {Preprint}, year = {2022}, author = {Keramati, Ramtin and Omer Gottesman and Celi, Leo Anthony} } @conference {chin_identifying_2022, title = {Identifying Structure in the MIMIC ICU Dataset}, year = {2022}, month = {dec}, publisher = {Proceedings of the Conference on Health, Inference, and Learning, 2022. }, organization = {Proceedings of the Conference on Health, Inference, and Learning, 2022. }, abstract = {The MIMIC-III dataset, containing trajectories of 40,000 ICU patients, is one of the most popular datasets in machine learning for health space. However, there has been very little systematic exploration to understand what is the natural structure of these data{\textendash}-most analyses enforce some type of top-down clustering or embedding. We take a bottom-up approach, identifying consistent structures that are robust across a range of embedding choices. We identified two dominant structures sorted by either fraction-inspired oxygen or creatinine {\textendash}- both of which were validated as the key features by our clinical co-author. Our bottom-up approach in studying the macro-structure of a dataset can also be adapted for other datasets.}, url = {https://openreview.net/forum?id=3vfn-cmUYQF}, author = {Chin, Zad and Raval, Shivam and Finale Doshi-Velez and Wattenberg, Martin and Celi, Leo Anthony} } @article {yacoby_if_2022, title = {{\textquotedblleft}If it didn{\textquoteright}t happen, why would I change my decision?{\textquotedblright}: How Judges Respond to Counterfactual Explanations for the Public Safety Assessment}, journal = {Proceedings of the AAAI Conference on Human Computation and Crowdsourcing}, volume = {10}, year = {2022}, month = {oct}, pages = {219{\textendash}230}, abstract = {Many researchers and policymakers have expressed excitement about algorithmic explanations enabling more fair and responsible decision-making. However, recent experimental studies have found that explanations do not always improve human use of algorithmic advice. In this study, we shed light on how people interpret and respond to counterfactual explanations (CFEs){\textendash}-explanations that show how a model{\textquoteright}s output would change with marginal changes to its input(s){\textendash}-in the context of pretrial risk assessment instruments (PRAIs). We ran think-aloud trials with eight sitting U.S. state court judges, providing them with recommendations from a PRAI that includes CFEs. We found that the CFEs did not alter the judges{\textquoteright} decisions. At first, judges misinterpreted the counterfactuals as real{\textendash}-rather than hypothetical{\textendash}-changes to defendants. Once judges understood what the counterfactuals meant, they ignored them, stating their role is only to make decisions regarding the actual defendant in question. The judges also expressed a mix of reasons for ignoring or following the advice of the PRAI without CFEs. These results add to the literature detailing the unexpected ways in which people respond to algorithms and explanations. They also highlight new challenges associated with improving human-algorithm collaborations through explanations.}, keywords = {CAREER, IIS-1750358}, issn = {2769-1349}, doi = {10.1609/hcomp.v10i1.22001}, url = {https://ojs.aaai.org/index.php/HCOMP/article/view/22001}, author = {Yacoby, Yaniv and Ben Green and Jr, Christopher L. Griffin and Finale Doshi-Velez} } @article {zhang_interpretable_2022, title = {An interpretable RL framework for pre-deployment modeling in ICU hypotension management}, journal = {npj Digital Medicine}, volume = {5}, number = {1}, year = {2022}, note = {Number: 1 Publisher: Nature Publishing Group}, month = {nov}, pages = {1{\textendash}10}, abstract = {Computational methods from reinforcement learning have shown promise in inferring treatment strategies for hypotension management and other clinical decision-making challenges. Unfortunately, the resulting models are often difficult for clinicians to interpret, making clinical inspection and validation of these computationally derived strategies challenging in advance of deployment. In this work, we develop a general framework for identifying succinct sets of clinical contexts in which clinicians make very different treatment choices, tracing the effects of those choices, and inferring a set of recommendations for those specific contexts. By focusing on these few key decision points, our framework produces succinct, interpretable treatment strategies that can each be easily visualized and verified by clinical experts. This interrogation process allows clinicians to leverage the model{\textquoteright}s use of historical data in tandem with their own expertise to determine which recommendations are worth investigating further e.g. at the bedside. We demonstrate the value of this approach via application to hypotension management in the ICU, an area with critical implications for patient outcomes that lacks data-driven individualized treatment strategies; that said, our framework has broad implications on how to use computational methods to assist with decision-making challenges on a wide range of clinical domains.}, keywords = {CAREER, IIS-1750358}, issn = {2398-6352}, doi = {10.1038/s41746-022-00708-4}, url = {https://www.nature.com/articles/s41746-022-00708-4}, author = {Zhang, Kristine and Wang, Henry and Du, Jianzhun and Brian Chu and Robles Ar{\'e}valo, Aldo and Kindle, Ryan and Celi, Leo Anthony and Finale Doshi-Velez} } @conference {chiu_joint_2022, title = {A Joint Learning Approach for Semi-supervised Neural Topic Modeling}, booktitle = {Proceedings of the Sixth Workshop on Structured Prediction for NLP}, year = {2022}, month = {may}, pages = {40{\textendash}51}, publisher = {Association for Computational Linguistics}, organization = {Association for Computational Linguistics}, address = {Dublin, Ireland}, abstract = {Topic models are some of the most popular ways to represent textual data in an interpret-able manner. Recently, advances in deep generative models, specifically auto-encoding variational Bayes (AEVB), have led to the introduction of unsupervised neural topic models, which leverage deep generative models as opposed to traditional statistics-based topic models. We extend upon these neural topic models by introducing the Label-Indexed Neural Topic Model (LI-NTM), which is, to the extent of our knowledge, the first effective upstream semi-supervised neural topic model. We find that LI-NTM outperforms existing neural topic models in document reconstruction benchmarks, with the most notable results in low labeled data regimes and for data-sets with informative labels; furthermore, our jointly learned classifier outperforms baseline classifiers in ablation studies.}, keywords = {CAREER, IIS-1750358, To Do: CV}, doi = {10.18653/v1/2022.spnlp-1.5}, url = {https://aclanthology.org/2022.spnlp-1.5}, author = {Chiu, Jeffrey and Mittal, Rajat and Tumma, Neehal and Abhishek Sharma and Finale Doshi-Velez} } @conference {wu_learning_2022, title = {Learning Optimal Summaries of Clinical Time-series with Concept Bottleneck Models}, booktitle = {Proceedings of the 7th Machine Learning for Healthcare Conference}, year = {2022}, note = {ISSN: 2640-3498}, pages = {648{\textendash}672}, publisher = {PMLR}, organization = {PMLR}, abstract = {Despite machine learning models{\textquoteright} state-of-the-art performance in numerous clinical prediction and intervention tasks, their complex black-box processes pose a great barrier to their real-world deployment. Clinical experts must be able to understand the reasons behind a model{\textquoteright}s recommendation before taking action, as it is crucial to assess for criteria other than accuracy, such as trust, safety, fairness, and robustness. In this work, we enable human inspection of clinical timeseries prediction models by learning concepts, or groupings of features into high-level clinical ideas such as illness severity or kidney function. We also propose an optimization method which then selects the most important features within each concept, learning a collection of sparse prediction models that are sufficiently expressive for examination. On a real-world task of predicting vasopressor onset in ICU units, our algorithm achieves predictive performance comparable to state-of-the-art deep learning models while learning concise groupings conducive for clinical inspection.}, url = {https://proceedings.mlr.press/v182/wu22a.html}, author = {Wu, Carissa and Parbhoo, Sonali and Havasi, Marton and Finale Doshi-Velez} } @conference {tang_leveraging_2022, title = {Leveraging Factored Action Spaces for Efficient Offline Reinforcement Learning in Healthcare}, booktitle = {Decision Awareness in Reinforcement Learning Workshop at ICML 2022}, year = {2022}, month = {jul}, abstract = {Many reinforcement learning (RL) applications have combinatorial action spaces, where each action is a composition of sub-actions. A standard RL approach ignores this inherent factorization structure, resulting in a potential failure to make meaningful inferences about rarely observed sub-action combinations; this is particularly problematic for offline settings, where data may be limited. In this work, we propose a form of linear Q-function decomposition induced by factored action spaces. We study the theoretical properties of our approach, identifying scenarios where it is guaranteed to lead to zero bias when used to approximate the Q-function. Outside the regimes with theoretical guarantees, we show that our approach can still be useful because it leads to better sample efficiency without necessarily sacrificing policy optimality, allowing us to achieve a better bias-variance trade-off. Across several offline RL problems using simulators and real-world datasets motivated by healthcare problems, we demonstrate that incorporating factored action spaces into value-based RL can result in better-performing policies. Our approach can help an agent make more accurate inferences within under-explored regions of the state-action space when applying RL to observational datasets.}, keywords = {CV - Journals}, url = {https://openreview.net/forum?id=wl_o_hilncS}, author = {Tang, Shengpu and Maggie Makar and Sjoding, Michael and Finale Doshi-Velez and Wiens, Jenna} } @article {yacoby_mitigating_nodate, title = {Mitigating the Effects of Non-Identifiability on Inference for Bayesian Neural Networks with Latent Variables}, journal = {Journal of Machine Learning Research 23 (2022)}, year = {2022}, abstract = {Bayesian Neural Networks with Latent Variables (BNN+LVs) capture predictive uncertainty by explicitly modeling model uncertainty (via priors on network weights) and environmental stochasticity (via a latent input noise variable). In this work, we first show that BNN+LV suffers from a serious form of non-identifiability: explanatory power can be transferred between the model parameters and latent variables while fitting the data equally well. We demonstrate that as a result, in the limit of infinite data, the posterior mode over the network weights and latent variables is asymptotically biased away from the ground-truth. Due to this asymptotic bias, traditional inference methods may in practice yield parameters that generalize poorly and misestimate uncertainty. Next, we develop a novel inference procedure that explicitly mitigates the effects of likelihood non-identifiability during training and yields high-quality predictions as well as uncertainty estimates. We demonstrate that our inference method improves upon benchmark methods across a range of synthetic and real data-sets.}, author = {Yacoby, Yaniv and Weiwei Pan and Finale Doshi-Velez} } @article {yao_policy_2022, title = {Policy Optimization with Sparse Global Contrastive Explanations}, journal = {Preprint}, year = {2022}, note = {arXiv:2207.06269 [cs]}, publisher = {arXiv}, abstract = {We develop a Reinforcement Learning (RL) framework for improving an existing behavior policy via sparse, user-interpretable changes. Our goal is to make minimal changes while gaining as much benefit as possible. We define a minimal change as having a sparse, global contrastive explanation between the original and proposed policy. We improve the current policy with the constraint of keeping that global contrastive explanation short. We demonstrate our framework with a discrete MDP and a continuous 2D navigation domain.}, doi = {10.48550/arXiv.2207.06269}, url = {http://arxiv.org/abs/2207.06269}, author = {Jiayu Yao and Parbhoo, Sonali and Weiwei Pan and Finale Doshi-Velez} } @article {penrod_success_2022, title = {Success of Uncertainty-Aware Deep Models Depends on Data Manifold Geometry}, journal = {Preprint}, year = {2022}, note = {arXiv:2208.01705 [cs]}, month = {aug}, publisher = {arXiv}, abstract = {For responsible decision making in safety-critical settings, machine learning models must effectively detect and process edge-case data. Although existing works show that predictive uncertainty is useful for these tasks, it is not evident from literature which uncertainty-aware models are best suited for a given dataset. Thus, we compare six uncertainty-aware deep learning models on a set of edge-case tasks: robustness to adversarial attacks as well as out-of-distribution and adversarial detection. We find that the geometry of the data sub-manifold is an important factor in determining the success of various models. Our finding suggests an interesting direction in the study of uncertainty-aware deep learning models.}, doi = {10.48550/arXiv.2208.01705}, url = {http://arxiv.org/abs/2208.01705}, author = {Penrod, Mark and Termotto, Harrison and Reddy, Varshini and Jiayu Yao and Finale Doshi-Velez and Weiwei Pan} } @proceedings {noauthor_towards_nodate, title = {Towards Robust Off-Policy Evaluation via Human Inputs}, journal = {Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society}, year = {2022}, url = {https://dl.acm.org/doi/abs/10.1145/3514094.3534198?casa_token=QYwH8TIC-QYAAAAA:9bP2zidmphQac-O4CP0-uibVvKm-wFR4QiQNQsSKwTT20NGPnrncrxS_ZVoZOl3xnWcnrWopRpEH}, author = {Singh, Harvineet and Joshi, Shalmali and Finale Doshi-Velez and Himabindu Lakkaraju} } @conference {subhash_what_2022, title = {What Makes a Good Explanation?: A Harmonized View of Properties of Explanations}, booktitle = {Progress and Challenges in Building Trustworthy Embodied AI}, year = {2022}, month = {nov}, abstract = {Interpretability provides a means for humans to verify aspects of machine learning (ML) models. Different tasks require explanations with different properties. However, presently, there is a lack of standardization in assessing properties of explanations: different papers use the same term to mean different quantities, and different terms to mean the same quantity. This lack of standardization prevents us from rigorously comparing explanation systems. In this work, we survey explanation properties defined in the current interpretable ML literature, we synthesize properties based on what they measure, and describe the trade-offs between different formulations of these properties. We provide a unifying framework for comparing properties of interpretable ML.}, keywords = {CAREER, Duplicate Presentation, IIS-1750358}, url = {https://openreview.net/forum?id=YDyLZWwpBK2}, author = {Subhash, Varshini and Chen, Zixi and Havasi, Marton and Weiwei Pan and Finale Doshi-Velez} } @conference {chen_what_2022, title = {What Makes a Good Explanation?: A Harmonized View of Properties of Explanations}, booktitle = {Workshop on Trustworthy and Socially Responsible Machine Learning, NeurIPS 2022}, year = {2022}, month = {nov}, abstract = {Interpretability provides a means for humans to verify aspects of machine learning (ML) models. Different tasks require explanations with different properties. However, presently, there is a lack of standardization in assessing properties of explanations: different papers use the same term to mean different quantities, and different terms to mean the same quantity. This lack of standardization prevents us from rigorously comparing explanation systems. In this work, we survey explanation properties defined in the current interpretable ML literature, we synthesize properties based on what they measure, and describe the trade-offs between different formulations of these properties. We provide a unifying framework for comparing properties of interpretable ML.}, keywords = {IIS-1750358}, url = {https://openreview.net/forum?id=TnFHizNosji}, author = {Chen, Zixi and Subhash, Varshini and Havasi, Marton and Weiwei Pan and Finale Doshi-Velez} } @article {narayanan_when_2022, title = {(When) Are Contrastive Explanations of Reinforcement Learning Helpful?}, journal = {Preprint}, year = {2022}, note = {arXiv:2211.07719 [cs]}, publisher = {arXiv}, abstract = {Global explanations of a reinforcement learning (RL) agent{\textquoteright}s expected behavior can make it safer to deploy. However, such explanations are often difficult to understand because of the complicated nature of many RL policies. Effective human explanations are often contrastive, referencing a known contrast (policy) to reduce redundancy. At the same time, these explanations also require the additional effort of referencing that contrast when evaluating an explanation. We conduct a user study to understand whether and when contrastive explanations might be preferable to complete explanations that do not require referencing a contrast. We find that complete explanations are generally more effective when they are the same size or smaller than a contrastive explanation of the same policy, and no worse when they are larger. This suggests that contrastive explanations are not sufficient to solve the problem of effectively explaining reinforcement learning policies, and require additional careful study for use in this context.}, doi = {10.48550/arXiv.2211.07719}, url = {http://arxiv.org/abs/2211.07719}, author = {Narayanan, Sanjana and Isaac Lage and Finale Doshi-Velez} } @conference {coker_wide_2022, title = {Wide Mean-Field Bayesian Neural Networks Ignore the Data}, booktitle = {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics}, year = {2022}, note = {ISSN: 2640-3498}, pages = {5276{\textendash}5333}, publisher = {PMLR}, organization = {PMLR}, abstract = {Bayesian neural networks (BNNs) combine the expressive power of deep learning with the advantages of Bayesian formalism. In recent years, the analysis of wide, deep BNNs has provided theoretical insight into their priors and posteriors. However, we have no analogous insight into their posteriors under approximate inference. In this work, we show that mean-field variational inference entirely fails to model the data when the network width is large and the activation function is odd. Specifically, for fully-connected BNNs with odd activation functions and a homoscedastic Gaussian likelihood, we show that the optimal mean-field variational posterior predictive (i.e., function space) distribution converges to the prior predictive distribution as the width tends to infinity. We generalize aspects of this result to other likelihoods. Our theoretical results are suggestive of underfitting behavior previously observered in BNNs. While our convergence bounds are non-asymptotic and constants in our analysis can be computed, they are currently too loose to be applicable in standard training regimes. Finally, we show that the optimal approximate posterior need not tend to the prior if the activation function is not odd, showing that our statements cannot be generalized arbitrarily.}, url = {https://proceedings.mlr.press/v151/coker22a.html}, author = {Beau Coker and Bruinsma, Wessel P. and Burt, David R. and Weiwei Pan and Finale Doshi-Velez} } @proceedings {692026, title = {"If it didn{\textquoteright}t happen, why would I change my decision?": How Judges Respond to Counterfactual Explanations for the Public Safety Assessment"}, journal = {proceeding at the Human Centered Explainable AI Conference: CHI Workshop on Human Centered Explainable AI (HCXAI}, volume = {2}, year = {2022}, pages = {1-24}, author = {Yacoby, Y. and B. Green and C. Griffin, Jr. and Doshi-Velez, F.} } @conference {692023, title = {Identification of Subgroups With Similar Benefits in Off-Policy Policy Evaluation}, booktitle = {Proceedings of the Conference on Health, Inference, and Learning}, year = {2022}, pages = {397-410}, author = {R. Keramati and Gottesman, O. and Celi, L. and Doshi-Velez, F. and Brunskill, E.} } @conference {wang_learning_2021, title = {Learning MDPs from Features: Predict-Then-Optimize for Sequential Decision Making by Reinforcement Learning}, booktitle = {Advances in Neural Information Processing Systems}, volume = {34}, year = {2021}, pages = {8795{\textendash}8806}, publisher = {Curran Associates, Inc.}, organization = {Curran Associates, Inc.}, abstract = {In the predict-then-optimize framework, the objective is to train a predictive model, mapping from environment features to parameters of an optimization problem, which maximizes decision quality when the optimization is subsequently solved. Recent work on decision-focused learning shows that embedding the optimization problem in the training pipeline can improve decision quality and help generalize better to unseen tasks compared to relying on an intermediate loss function for evaluating prediction quality. We study the predict-then-optimize framework in the context of sequential decision problems (formulated as MDPs) that are solved via reinforcement learning. In particular, we are given environment features and a set of trajectories from training MDPs, which we use to train a predictive model that generalizes to unseen test MDPs without trajectories. Two significant computational challenges arise in applying decision-focused learning to MDPs: (i) large state and action spaces make it infeasible for existing techniques to differentiate through MDP problems, and (ii) the high-dimensional policy space, as parameterized by a neural network, makes differentiating through a policy expensive. We resolve the first challenge by sampling provably unbiased derivatives to approximate and differentiate through optimality conditions, and the second challenge by using a low-rank approximation to the high-dimensional sample-based derivatives. We implement both Bellman-based and policy gradient-based decision-focused learning on three different MDP problems with missing parameters, and show that decision-focused learning performs better in generalization to unseen tasks.}, keywords = {MURI Grant Number W911NF-17-1-0370, MURI Grant Number W911NF-18-1-0208, NSF CAREER IIS-1750358}, url = {https://proceedings.neurips.cc/paper/2021/hash/49e863b146f3b5470ee222ee84669b1c-Abstract.html}, author = {Wang, Kai and Shah, Sanket and Chen, Haipeng and Perrault, Andrew and Finale Doshi-Velez and Milind Tambe} } @conference {692025, title = {State Relevance for Off-Policy Evaluation}, booktitle = {proceeding at the International Conference on Machine Learning}, volume = {1}, year = {2021}, pages = {1-20}, url = {arXiv:2109.06310v1}, author = {S. Shein and Ma, Y. and Gottesman, O. and Doshi-Velez, F.} } @conference {692024, title = {Power Constrained Bandits}, booktitle = { proceedings at the International Conference on Machine Learning for Healthcare }, volume = {4}, year = {2021}, pages = {1-50}, author = {Futoma, J. and M. Simons and Doshi-Velez, F. and Kamaleswaran, R.} } @proceedings {684013, title = {Preferential Mixture-of-Experts: Interpretable Models that Rely on Human Expertise as Much as Possible}, journal = {proceeding at the Clinical Research Informatics American Medical Informatics Association Summit (AMIA), }, volume = {1}, year = {2021}, pages = {1-15}, author = {Pradier, M. and J. Zazo and S. Parbhoo and Perlis, R. and Zazzi, M. and Doshi-Velez, F.} } @article {684012, title = {How machine-learning recommendations influence clinician treatment selections: the example of antidepressant selection}, journal = {Translational psychiatry}, volume = {1}, year = {2021}, pages = {1-9}, author = {Jacobs, M. and Pradier, M. and McCoy, T. and Perlis, R. and Doshi-Velez, F. and Gajos, K.} } @proceedings {684011, title = {Benchmarks, Algorithms, and Metrics for Hierarchical Disentanglement}, journal = {Proceeding at the International Conference on Machine Learning (ICML)}, volume = {2}, year = {2021}, pages = {1-23}, author = {Ross, A. and Doshi-Velez, F.} } @article {684010, title = {Depth-Bounded Statistical PCFG Induction as a Model of Human Grammar Acquisition}, journal = {Computational Linguistics}, volume = {47}, number = {1}, year = {2021}, pages = {1-36}, author = {L. Jin and L. Schwartz and Doshi-Velez, F. and Miller, T. and W. Schuler} } @article {684009, title = {Learning Under Adversarial and Interventional Shifts}, journal = {Preprint}, volume = {1}, year = {2021}, pages = {1-19}, author = {H. Singh and S. Joshi and Doshi-Velez, F. and Lakkaraju, H.} } @article {684008, title = {Machine Learning Techniques for Accountability}, journal = {AI Magazine}, volume = {42}, number = {1}, year = {2021}, pages = {1}, author = {B. Kim and Doshi-Velez, F.} } @proceedings {684007, title = {Designing AI for Trust and Collaboration in Time-Constrained Medical Decisions: A Sociotechnical Lens}, journal = {proceedings at the Conference on Human Factors in Computing Systems (CHI)}, volume = {2}, year = {2021}, pages = {1-14}, author = {Jacobs, M. and J. He and M F. Pradier and B. Lam and A. Ahn and McCoy, T. and Perlis, R. and Doshi-Velez, F. and Gajos, K.} } @proceedings {684006, title = {Designing AI for Trust and Collaboration in Time-Constrained Medical Decisions: A Sociotechnical Lens}, journal = {proceedings at the Conference on Human Factors in Computing Systems (CHI)}, volume = {2}, year = {2021}, pages = {1-14}, author = {Jacobs, M. and J. He and M F. Pradier and B. Lam and A. Ahn and McCoy, T. and Perlis, R. and Doshi-Velez, F. and Gajos, K.} } @article {684004, title = {Power Constrained Bandit}, journal = {proceedings at the Machine Learning for Healthcare Conference}, volume = {4}, year = {2021}, pages = {1-50}, url = {https://static1.squarespace.com/static/59d5ac1780bd5ef9c396eda6/t/60fb3a37b5d5c73c4aba9c58/1627077177096/Power_Constrained_Bandits_MLHC_2021.pdf}, author = {Yao, J. and Brunskill, E. and Pan, W. and S. Murphy and Doshi-Velez, F.} } @article {684003, title = {Generalization in Clinical Prediction Models: The Blessing and Curse of Measurement Indicator Variables}, journal = {Critical Care Explorations}, volume = {1}, year = {2021}, pages = {1-11}, author = {Futoma, J. and M. Simons and Doshi-Velez, F. and Kamaleswaran, R.} } @proceedings {684001, title = {Prediction-focused Mixture Models }, journal = {proceeding at the International Conference on Machine Learning: Information-Theoretic Methods for Rigorous, Responsible, and Reliable Machine Learning (ICML), 2021}, volume = {1}, year = {2021}, pages = {1-9}, author = {Narayanan, S. and Sharma, A. and Zeng, C. and Doshi-Velez, F.} } @proceedings {683999, title = {Wide Mean-Field Variational Bayesian Neural Networks Ignore the Data}, journal = {proceedings at the International Conference on Machine Learning: Workshop on Uncertainty \& Robustness in Deep Learning (ICML)}, volume = {1}, year = {2021}, pages = {1-15}, author = {Coker, B. and S. Parbhoo and Doshi-Velez, F.} } @proceedings {683998, title = {Wide Mean-Field Variational Bayesian Neural Networks Ignore the Data}, journal = {proceedings at the International Conference on Machine Learning: Workshop on Uncertainty \& Robustness in Deep Learning (ICML)}, volume = {1}, year = {2021}, pages = {1-15}, author = {Coker, B. and S. Parbhoo and Doshi-Velez, F.} } @proceedings {683995, title = {Promises and Pitfalls of Black-Box Concept Learning Models}, journal = {proceeding at the International Conference on Machine Learning: Workshop on Theoretic Foundation, Criticism, and Application Trend of Explainable AI,}, volume = {1}, year = {2021}, pages = {1-13}, author = {A. Mahinpei and J. Clark and I. Lage and Doshi-Velez, F. and P. WeiWei} } @article {683994, title = {Optimizing for Interpretability in Deep Neural Networks with Tree Regularization}, journal = {Journal of AI Research (JAIR)}, volume = {1}, year = {2021}, pages = {1-37}, author = {Wu, M. and S. Parbhoo and Hughes, M. and V. Roth and Doshi-Velez, F.} } @proceedings {683993, title = {Pre-emptive Learning to Defer for Sequential Medical Decision-Making Under Uncertainty}, journal = {proceeding at the International Conference on Machine Learning: Workshop on Neglected Assumptions in Causal Inference (ICML)}, volume = {1}, year = {2021}, pages = {1-13}, author = {S. Parbhoo and S. Shalmali and Doshi-Velez, F.} } @proceedings {683992, title = {On formalizing causal off-policy evaluation for sequential decision-making}, journal = {proceeding at the International Conference on Machine Learning: Workshop on Neglected Assumptions in Causal Inference (ICML)}, year = {2021}, author = {Parbhoo, S and Shalmali, J and Doshi-Velez, F.} } @proceedings {683990, title = {Learning Predictive and Interpretable Timeseries Summaries from ICU Data}, journal = {proceeding at the Conference on American Medical Informatics Association (AMIA)}, volume = {1}, year = {2021}, pages = {1-10}, author = {Johnson, N and Parbhoo, S and Ross, A and Doshi-Velez, F.} } @proceedings {683989, title = {Learning Predictive and Interpretable Timeseries Summaries from ICU Data}, journal = {proceeding at the Conference on American Medical Informatics Association (AMIA)}, volume = {1}, year = {2021}, pages = {1-10}, author = {Johnson, N and Parbhoo, S and Ross, A and Doshi-Velez, F.} } @article {671659, title = {How machine learning recommendations influence clinician treatment selections: example of antidepressant selection}, journal = {Translational Psychiatry}, volume = {1}, year = {2021}, pages = {1-9}, author = {Jacobs, M. and Pradier, M. and McCoy, T. and P. Roy and Doshi-Velez, F. and G. Krzysztof} } @conference {671657, title = {Evaluating the Interpretability of Generative Models by Interactive Reconstruction}, booktitle = { proceeding at the Conference on Human Factors in Computing Systems (CHI), 2021}, volume = {1}, year = {2021}, pages = {1-20}, author = {Ross, A. and N. Chen and E. Hang and E. Glassman and Doshi-Velez, F.} } @conference {671656, title = {Designing AI for Trust in Collaboration in Time-Constrained Medical Decisions: A Sociotechnical Lens}, booktitle = {proceeding at the Conference on Human Factors in Computing Systems (CHI)}, volume = {1}, year = {2021}, pages = {1-14}, author = {Jacobs, M. and J. He and Pradier, M. and B. Lam and A. Ahn and McCoy, T. and Perlis, R. and Doshi-Velez, F. and Gajos, K.} } @conference {671649, title = {Identifying Decision Points for Safe and Interpretable Reinforcement Learning in Hypotension Treatment}, booktitle = {NeurIPS Workshop on Machine Learning for Health}, volume = {1}, year = {2021}, pages = {1-9}, author = {Zhang, K. and Wang, H. and Du, J. and B. Chu and Kindle, R. and Celi, L. and Doshi-Velez, F.} } @conference {671575, title = {Preferential Mixture-of-Experts: Interpretable Models that Rely on Human Expertise as Much as Possible}, booktitle = {American Medical Informatics Association (AMIA)}, volume = {1}, year = {2021}, pages = {1-13}, author = {Pradier, M. and J. Zazo and S. Parbhoo and Perlis, R. and Zazzi, M. and Doshi-Velez, F.} } @article {684052, title = {The myth of generalisability in clinical research and machine learning in health care}, journal = {The Lancet Digital Health}, volume = {3}, year = {2020}, pages = {1-19}, author = {Futoma, J. and M. Simons and T. Panch and Doshi-Velez, F. and Celi, L.} } @proceedings {684028, title = {Interpretable Off-Policy Evaluation in Reinforcement Learning by Highlighting Influential Transitions }, journal = {presented at the International Conference on Machine Learning}, volume = {3}, year = {2020}, pages = {1-19}, author = {Gottesman, O. and Futoma, J. and Liu, Y. and S. Parbhoo and Celi, L. and Brunskill, E. and Doshi-Velez, F.} } @article {684023, title = {Learning Interpretable Concept-Based Models with Human Feedback}, journal = {presented at the International Conference on Machine Learning: Workshop on Human Interpretability in Machine Learnin}, volume = {1}, year = {2020}, pages = {1-11}, author = {I. Lage and Doshi-Velez, F.} } @article {684022, title = {Learning Interpretable Concept-Based Models with Human Feedback}, journal = {presented at the International Conference on Machine Learning: Workshop on Human Interpretability in Machine Learnin}, volume = {1}, year = {2020}, pages = {1-11}, author = {I. Lage and Doshi-Velez, F.} } @article {684021, title = {Artificial Intelligence \& Cooperation}, journal = {Preprint}, volume = {1}, year = {2020}, pages = {1-4}, author = {E. Bertino and Doshi-Velez, F. and M. Gini and D. Lopresti and D. Parkes} } @article {684014, title = {Predicting change in diagnosis from major depression to bipolar disorder after antidepressant initiation}, journal = {Neuropsychopharmacology}, volume = {1}, year = {2020}, pages = {1-7}, author = {Pradier, M. and C. Hughes and McCoy, T. and S. Barroilhet and Doshi-Velez, F. and Perlis, R.} } @article {683996, title = {Big Data in the Assessment of Pediatric Medication Safety}, journal = {Pediatrics}, volume = {1}, year = {2020}, pages = {1-11}, url = {https://finale.seas.harvard.edu/files/finale/files/big_data_in_the_assessment_of_pediatric_medication_safety.pdf}, author = {McMahon, A. and Cooper, W. and J. Brown and Carleton, B. and Doshi-Velez, F. and Kohane, I. and Goldman, J. and Hoffman, M. and Kamaleswaran, R. and Sakiyama, M. and Sekine, S. and Sturkenboom, M. and Turner, M. and Califf, R.} } @article {683997, title = {Big Data in the Assessment of Pediatric Medication Safety}, journal = {Pediatrics}, volume = {1}, year = {2020}, pages = {1-11}, url = {https://finale.seas.harvard.edu/files/finale/files/big_data_in_the_assessment_of_pediatric_medication_safety.pdf}, author = {McMahon, A. and Cooper, W. and J. Brown and Carleton, B. and Doshi-Velez, F. and Kohane, I. and Goldman, J. and Hoffman, M. and Kamaleswaran, R. and Sakiyama, M. and Sekine, S. and Sturkenboom, M. and Turner, M. and Califf, R.} } @conference {671735, title = {A Bayesian Approach to Learning Bandit Structure in Markov Decision Processes}, booktitle = {proceeding at the Conference on Neural Information Processing Systems (NeurIPS): Workshop on Real World Reinforcement Learning}, year = {2020}, pages = {1-12}, author = {Zhang, K. and Gottesman, O. and Doshi-Velez, F.} } @conference {671734, title = {A Bayesian Approach to Learning Bandit Structure in Markov Decision Processes}, booktitle = {proceedings at the Conference on Neural Information Processing Systems (NeurIPS): Workshop on Real World Reinforcement Learning}, year = {2020}, pages = {1-12}, author = {Zhang, K. and Gottesman, O. and Doshi-Velez, F.} } @report {671730, title = {Artificial Intelligence and Cooperation}, year = {2020}, pages = {1-4}, edition = {1}, url = {arXiv:2012.06034}, author = {E. Bertino and Doshi-Velez, F. and M. Gini and D. Lopresti and D. Parkes} } @article {671652, title = {Assessment of a Prediction Model for Antidepressant Treatment Stability Using Supervised Topic Models }, journal = { JAMA Network Open}, year = {2020}, pages = {1-14}, author = {Hughes, M. and Pradier, M. and Ross, A. and M. McCoy and Perlis, R. and Doshi-Velez, F.} } @article {671594, title = {Incorporating Interpretable Output Constraints in Bayesian Neural Networks}, journal = {proceeding at the Conference on Neural Information Processing Systems (NeurIPS)}, volume = {2}, year = {2020}, pages = {1-17}, author = {W. Yang and Lorch, L. and Graule, M. and Lakkaraju, H. and Doshi-Velez, F.} } @article {671586, title = {Predicting change in diagnosis from major depression to bipolar disorder after antidepressant initiation}, journal = {Neuropsychopharmacology}, volume = {1}, year = {2020}, pages = {1-7}, author = {Pradier, M. and Hughes, M. and McCoy, T. and S. Barroilhet and Doshi-Velez, F. and Perlis, R.} } @conference {671571, title = {Shaping Control Variates for Off-Policy Evaluation}, booktitle = {NeurIPS Workshop on Offline Reinforcement Learning}, year = {2020}, pages = {1-9}, author = {S. Parbhoo and Gottesman, O. and Doshi-Velez, F.} } @article {671570, title = {The Myth of Generalizability in Clinical Research and Machine Learning in Healthcare}, journal = {Lancet Digital Health}, volume = {2}, year = {2020}, pages = {1-5}, url = {https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7444947/}, author = {Futoma, J. and M. Simons and T. Panch and Doshi-Velez, F. and Celi, L.} } @conference {671566, title = {Transfer Learning from Well-Curated to Less-Resourced Populations with HIV}, booktitle = {proceeding at the Machine Learning for Healthcare Conference}, year = {2020}, pages = {1-20}, author = {S. Parbhoo and M. Wieser and V. Roth and Doshi-Velez, F.} } @conference {671563, title = {Model-Based Reinforcement Learning for Semi-Markov Decision Processes with Neural ODEs.}, booktitle = {Neural Information Processing Systems Conference (NeurIPS) 2020}, volume = {2}, year = {2020}, pages = {1-21}, author = {Du, J. and Futoma, J. and Doshi-Velez, F.} } @article {661205, title = {Human-in-the-Loop Learning of Interpretable and Intuitive Representations}, journal = {ICML Workshop on Human Interpretability in Machine Learning, }, volume = {1}, year = {2020}, pages = {1-10}, author = {I. Lage and Doshi-Velez, F.} } @article {660601, title = {Power-Constrained Bandits}, journal = {ICML Workshop on Theoretical Foundations of Reinforcement Learning}, volume = {2}, year = {2020}, pages = {1-30}, author = {Yao, J. and Brunskill, E. and Pan, W. and S. Murphy and Doshi-Velez, F.} } @article {660600, title = {PoRB-Nets: Poisson Process Radial Basis Function Networks}, journal = {UAI}, year = {2020}, pages = {1-59}, author = {Coker, B. and M. Fernandez-Pradier and Doshi-Velez, F.} } @article {660595, title = {PAC Imitation and Model-based Batch Learning of Contextual MDPs}, journal = {ICML Workshop on Theoretical Foundations of Reinforcement Learning}, volume = {2}, year = {2020}, pages = {1-21}, author = {Y. Nair and Doshi-Velez, F.} } @article {660594, title = {PAC Imitation and Model-based Batch Learning of Contextual MDPs}, journal = {ICML Workshop on Inductive Biases, Invariances and Generalization in RL}, volume = {2}, year = {2020}, pages = {1-21}, author = {Y. Nair and Doshi-Velez, F.} } @article {660590, title = {Learned Uncertainty-Aware (LUNA) Bases for Bayesian Regression using Multi-Headed Auxiliary Networks}, journal = {ICML Workshop on Uncertainty in Deep Learning}, volume = {2}, year = {2020}, pages = {1-18}, author = {S. Thakur and C. Lorsung and Yacoby, Y. and Doshi-Velez, F. and Pan, W.} } @article {660589, title = {Is Deep Reinforcement Learning Ready for Practical Applications in Healthcare? A Sensitivity Analysis of Duel-DDQN for Sepsis Treatment}, journal = {AMIA}, volume = {1}, year = {2020}, pages = {1-13}, author = {M. Lu and Z. Shahn and D. Sow and Doshi-Velez, F. and L. Lehman} } @conference {660552, title = {Interpretable Off-Policy Evaluation in Reinforcement Learning by Highlighting Influential Transitions}, booktitle = {International Conference on Machine Learning}, volume = {2}, year = {2020}, pages = {1-17}, author = {Gottesman, O. and Futoma, J. and Liu, Y. and S. Parbhoo and LA. Celi and Brunskill, E. and Doshi-Velez, F.} } @article {660547, title = {Interpretable Off-Policy Evaluation in Reinforcement Learning by Highlighting Influential Transitions}, journal = {International Conference on Machine Learning (IMCL).}, volume = {2}, year = {2020}, pages = {1-17}, author = {Gottesman, O. and Futoma, J. and Liu, Y. and S. Parbhoo and LA. Celi and Brunskill, E. and Doshi-Velez, F.} } @article {660544, title = {Failures of Variational Autoencoders and their Effects on Downstream Tasks}, journal = {ICML Workshop on Uncertainty in Deep Learning}, volume = {1}, year = {2020}, pages = {1-39}, author = {Yacoby, Y. and Pan, W. and Doshi-Velez, F.} } @article {660541, title = {Discussions on Horseshoe Regularisation for Machine Learning in Complex and Deep Models}, journal = {International Statistical Review}, volume = {1}, year = {2020}, pages = {1-3}, author = {Ghosh, S. and Doshi-Velez, F.} } @article {660539, title = {CRUDS: Counterfactual Recourse Using Disentangled Subspaces}, journal = {ICML Workshop on Human Interpretability in Machine Learning}, year = {2020}, pages = {1-23}, author = {M. Downs, and J. Chu, and Yacoby, Y. and Doshi-Velez, F. and P. WeiWei} } @article {660537, title = {BaCOUn: Bayesian Classifers with Out-of-Distribution Uncertainty}, journal = {ICML Workshop on Uncertainty in Deep Learning}, volume = {1}, year = {2020}, pages = {1-24}, author = {T. Guenais and D. Vamvourellis and Yacoby, Y. and Doshi-Velez, F. and Pan, W.} } @article {660535, title = {Amortised Variational Inference for Hierarchical Mixture Models}, journal = {ICML Workshop on Uncertainty in Deep Learning}, year = {2020}, pages = {1-11}, author = {J. Antoran, and Yao, J. and Pan, W. and Doshi-Velez, F. and J. Hernandez-Lobato} } @article {660532, title = {Active Screening on Recurrent Diseases Contact Networks with Uncertainty: a Reinforcement Learning Approach}, journal = {AAMAS Workshop on Multi-Agent Based Simulation (AAMAS)}, year = {2020}, pages = {1-12}, author = {HC. Ou, and Wang, K. and Doshi-Velez, F. and M. Tambe} } @article {651081, title = {Characterizing and Avoiding Problematic Global Optima of Variational Autoencoders}, journal = {Advances in Approximate Bayesian Inference}, volume = {1}, year = {2020}, pages = {1-17}, author = {Yacoby, Y. and Pan, W. and Doshi-Velez, F.} } @article {651007, title = {Defining Admissible Rewards for High-Confidence Policy Evaluation in Batch Reinforcement Learning}, journal = {ACM Conference on Health, Inference and Learning}, volume = {2}, year = {2020}, pages = {1-9}, author = {Prasad, N. and Engelhardt, B. and Doshi-Velez, F.} } @article {651003, title = {Prediction Focused Topic Models via Feature Selection}, journal = {AISTATS}, volume = {2}, year = {2020}, pages = {1-19}, author = {J. Ren and Kunes, R. and Doshi-Velez, F.} } @article {651002, title = {POPCORN: Partially Observed Prediction Constrained Reinforcement Learning}, journal = {AISTATS}, volume = {2}, year = {2020}, pages = {1-18}, author = {Futoma, J. and Hughes, M. and Doshi-Velez, F.} } @article {650997, title = {Regional Tree Regularization for Interpretability in Deep Neural Networks}, journal = {AAAI}, volume = {3}, year = {2020}, pages = {1-9}, author = {Wu, M. and S. Parbhoo and Hughes, M. and Kindle, R. and Celi, L. and Zazzi, M. and Volker, R. and Doshi-Velez, F.} } @article {650998, title = {Regional Tree Regularization for Interpretability in Deep Neural Networks}, journal = {AAAI}, volume = {3}, year = {2020}, pages = {1-9}, author = {Wu, M. and S. Parbhoo and Hughes, M. and Kindle, R. and Celi, L. and Zazzi, M. and Volker, R. and Doshi-Velez, F.} } @article {650996, title = {Ensembles of Locally Independent Prediction Models}, journal = {AAAI}, volume = {3}, year = {2020}, pages = {1-11}, author = {Ross, A. and Pan, W. and Celi, L. and Doshi-Velez, F.} } @article {650995, title = {Identifying Distinct, Effective Treatments for Acute Hypotension with SODA-RL: Safely Optimized Diverse Accurate Reinforcement Learning}, journal = {AMIA CRI}, volume = {1}, year = {2020}, pages = {1-24}, author = {Futoma, F. and Masgood, M. and Doshi-Velez, F.} } @article {650994, title = {Interpretable Batch IRL to extract clinician goals in ICU HypotensionManagement}, journal = {AMIA CRI}, year = {2020}, pages = {636-645}, author = {Srinivasan, S. and Doshi-Velez, F.} } @article {650993, title = {Big Data in the Assessment of Pediatric Medication Safety}, journal = {Pediatrics}, volume = {145}, number = {2}, year = {2020}, pages = {1-11}, author = {McMahon, A. and Cooper, W. and J. Brown and Carleton, B. and Doshi-Velez, F. and Kohane, I. and Goldman, J. and Hoffman, M. and Kamaleswaran, R. and Sakiyama, M. and Sekine, S. and Sturkenboom, M. and Turner, M. and Califf, R.} } @article {650992, title = {Evaluating Machine Learning Articles}, journal = {JAMA}, volume = {322}, number = {18}, year = {2020}, pages = {1777-1779}, author = {Doshi-Velez, F. and Perlis, R.} } @article {650991, title = {Predicting treatment dropout after antidepressant initiation}, journal = {Translational Psychiatry}, volume = {10}, number = {1}, year = {2020}, pages = {1-8}, author = {Pradier, M. and McCoy, T. and Hughes, M. and Perlis, R. and Doshi-Velez, F.} } @article {wiens_no_2019, title = {Do no harm: a roadmap for responsible machine learning for health care}, journal = {Nature Medicine}, volume = {25}, number = {9}, year = {2019}, note = {Number: 9 Publisher: Nature Publishing Group}, pages = {1337{\textendash}1340}, abstract = {Interest in machine-learning applications within medicine has been growing, but few studies have progressed to deployment in patient care. We present a framework, context and ultimately guidelines for accelerating the translation of machine-learning-based interventions in health care. To be successful, translation will require a team of engaged stakeholders and a systematic process from beginning (problem formulation) to end (widespread deployment).}, keywords = {Health care, Research management, Scientific community}, issn = {1546-170X}, doi = {10.1038/s41591-019-0548-6}, url = {https://www.nature.com/articles/s41591-019-0548-6}, author = {Wiens, Jenna and Saria, Suchi and Sendak, Mark and Marzyeh Ghassemi and Liu, Vincent X. and Finale Doshi-Velez and Jung, Kenneth and Heller, Katherine and Kale, David and Saeed, Mohammed and Ossorio, Pilar N. and Thadaney-Israni, Sonoo and Goldenberg, Anna} } @conference {661204, title = {Towards Expressive Priors for Bayesian Neural Networks: Poisson Process Radial Basis Function Networks}, booktitle = { proceedings at The Conference on Uncertainty in Artificial Intelligence (UAI)}, volume = {1}, year = {2019}, pages = {1-37}, author = {Coker, B. and Pradier, M. and Doshi-Velez, F.} } @article {651140, title = {A Particle-Based Variational Approach to Bayesian Non-negative Matrix Factorization}, journal = {Journal of Machine Learning Research}, volume = {20}, number = {90}, year = {2019}, pages = {1-56}, author = {M. Masood and Doshi-Velez, F.} } @article {651129, title = {Model Selection in Bayesian Neural Networks via Horseshoe Priors}, journal = {Journal of Machine Learning Research}, volume = {20}, number = {182}, year = {2019}, pages = {1-46}, author = {Ghosh, S. and Yao, J. and Doshi-Velez, F.} } @article {651109, title = {Defining Admissible Rewards for High Confidence Policy Evaluation}, journal = {NeurIPS Workshop on Safety and Robustness in Decision-Making, }, volume = {1}, year = {2019}, pages = {1-12}, author = {Prasad, N. and Engelhardt, B. and Doshi-Velez, F.} } @article {651097, title = {Controlled Direct Effect Priors for Bayesian Neural Networks}, journal = {NeurIPS Workshop on Bayesian Deep Learning}, volume = {1}, year = {2019}, pages = {1-8}, author = {Ross, A. and Du, J. and Sharvit, Y. and Doshi-Velez, F.} } @article {651091, title = {Integrating AI Recommendations into The Pharmacologic Management of Major Depressive Disorder}, journal = {CSCW Workshop: Identifying Challenges and Opportunities in Human{\textendash}AI Collaboration in Healthcare}, volume = {1}, year = {2019}, pages = {1-5}, author = {Jacobs, M. and Perlis, R. and Pradier, M. and Doshi-Velez, F. and Mynatt, E. and Gajos, K.} } @article {651084, title = {Prediction Focused Topic Models Via Vocab Filtering}, journal = {NeurIPS Workshop on Human-Centric ML}, volume = {1}, year = {2019}, pages = {1-12}, author = {J. Ren and Russell, R. and Doshi-Velez, F.} } @article {651080, title = {Challenges in Computing and Optimizing Upper Bounds of Marginal Likelihood based on Chi-Square Divergences}, journal = {Advances in Approximate Bayesian Inference}, volume = {1}, year = {2019}, pages = {1-11}, author = {Pradier, M. and Hughes, M. and Doshi-Velez, F.} } @article {651079, title = {Projected BNNs: Avoiding Weight-space Pathologies by Learning Latent Representations of Neural Network Weights}, journal = {ACML Workshop on Weakly Supervised Learning Workshop}, volume = {3}, year = {2019}, pages = {1-15}, author = {Pradier, M. and Pan, W. and Yao, J. and Ghosh, S. and Doshi-Velez, F.} } @article {651076, title = {Prediction Focused Topic Models for Electronic Health Records}, journal = {NeurIPS Workshop on Machine Learning for Health}, volume = {1}, year = {2019}, pages = {1-13}, author = {J. Ren and Kunes, R. and Doshi-Velez, F.} } @article {651011, title = {Do no harm: A roadmap for responsible machine learning for healthcare}, journal = {Nature Medicine}, volume = {25}, number = {10}, year = {2019}, pages = {1337-1340}, author = {Wiens, J. and Saria, S. and Sendak, M. and Ghassemi, M. and Liu, V. and Doshi-Velez, F. and Jung, K. and Heller, K. and Kale, D. and Saeed, M. and Ossorio, P. and Thadaney-Israni, S. and Goldenberg, A.} } @article {651008, title = {Summarizing Agent Strategies}, journal = {Journal of Autonomous Agents and Multi-Agent Systems (AAMAS)}, volume = {33}, year = {2019}, pages = {628-644}, author = {Amir, O. and Doshi-Velez, F. and Sarne, D.} } @conference {650941, title = {The Application of Machine Learning Methods to Evaluate Predictors for Live Birth in Programmed Thaw Cycles}, booktitle = {in proceedings at the American Society for Reproductive Medicine Scientific Congress \& Expo (ASRM)}, year = {2019}, author = {Vaughan, D. and Pan, W. and Yacoby, Y. and Seidler, E. and Leung, A. and Doshi-Velez, F. and Sakkas, D.} } @conference {650940, title = {Output-Constrained Bayesian Neural Network}, booktitle = {proceedings at the International Conference on Machine Learning: Workshop on Understanding and Improving Generalization in Deep Learning(ICML)}, year = {2019}, author = {W. Yang and Lorch, L. and Graule, M. and Srinivasan, S. and Suresh, A. and Yao, J. and Pradier, M. and Doshi-Velez, F.} } @conference {650938, title = {Output-Constrained Bayesian Neural Networks}, booktitle = {proceedings at the International Conference on Machine Learning: Workshop on Uncertainty \& Robustness in Deep Learning (ICML)}, year = {2019}, author = {W. Yang and Lorch, L. and Graule, M. and Srinivasan, S. and Suresh, A. and Yao, J. and Pradier, M. and Doshi-Velez, F.} } @conference {650936, title = {Mitigating Model Non-Identifiability in BNN with Latent Variables}, booktitle = {proceedings at the International Conference on Machine Learning: Workshop on Uncertainty \& Robustness in Deep Learning (ICML)}, year = {2019}, author = {Yacoby, Y. and Pan, W. and Doshi-Velez, F.} } @conference {650935, title = {Quality of Uncertainty Quantification for Bayesian Neural Network Inference}, booktitle = {proceedings at the International Conference on Machine Learning: Workshop on Uncertainty \& Robustness in Deep Learning (ICML)}, year = {2019}, author = {Yao, J. and Pan, W. and Ghosh, S. and Doshi-Velez, F.} } @conference {650934, title = {Poisson Process Bayesian Neural Networks}, booktitle = {proceedings at the International Conference on Bayesian Nonparametrics (BNP)}, year = {2019}, author = {Coker, B. and Pradier, M. and Doshi-Velez, F.} } @conference {650930, title = {Toward Robust Policy Summarization}, booktitle = {proceedings at the International Joint Conference on Artificial Intelligence (IJCAI)}, year = {2019}, author = {I. Lage and Lifschitz, D. and Doshi-Velez, F. and Amir, O.} } @conference {650929, title = {Toward Robust Summarization of Agent Policies}, booktitle = {proceedings at the International Conference on Autonomous Agents and Multiagent Systems (AAMAS)}, year = {2019}, author = {I. Lage and Lifschitz, D. and Doshi-Velez, F. and Amir, O.} } @conference {650903, title = {Human Evaluation of Models Built for Interpretability}, booktitle = {proceedings at the 7th AAAI Conference on Human Computation and Crowdsourcing (HCOMP)}, year = {2019}, author = {I. Lage and E. Chen and J. He and M. Narayanan and B. Kim and S. Gershman and Doshi-Velez, F.} } @conference {650902, title = {Truly Batch Apprenticeship Learning with Deep Successor Features}, booktitle = {proceedings at the International Joint Conference on Artificial Intelligence (IJCAI)}, year = {2019}, author = {Srinivasan, S. and D. Lee and Doshi-Velez, F.} } @conference {650901, title = {Exploring Computational User Models for Agent Policy Summarization}, booktitle = {proceedings at the International Joint Conference on Artificial Intelligence: Workshop on Explainable Artificial Intelligence (IJCAI), }, year = {2019}, author = {Lage, I and Lifschitz, D. and Doshi-Velez, F. and Amir, O.} } @conference {650899, title = {Explainable Reinforcement Learning via Reward Decomposition}, booktitle = {in proceedings at the International Joint Conference on Artificial Intelligence. A Workshop on Explainable Artificial Intelligence.}, year = {2019}, author = {Z. Juozapaitis and A. Koul and A. Fern and M. Erwig and Doshi-Velez, F.} } @conference {650898, title = {Diversity-Inducing Policy Gradient: Using Maximum Mean Discrepancy to Find a Set of Diverse Policies}, booktitle = {proceedings at the International Joint Conference on Artificial Intelligence (IJCAI)}, year = {2019}, author = {M. Masood and Doshi Velez, F.} } @conference {650896, title = {Combining Parametric and Nonparametric Models for off-policy evaluation}, booktitle = {International Conference on Machine Learning (IMCL).}, year = {2019}, author = {Gottesman, O. and Liu, Y. and Susser, E. and Brunskill, E. and Doshi-Velez, F.} } @article {650619, title = {Assessing topic model relevance: Evaluation and informative priors}, journal = {Statistical Analysis and Data Mining}, volume = {12}, year = {2019}, pages = {210-222}, author = {Angela Fan and Finale Doshi-Velez and Miratrix, Luke} } @article {630307, title = {Guidelines for reinforcement learning in healthcare}, journal = {Nature Medicine}, volume = {25}, year = {2019}, pages = {16-18}, author = {Omer Gottesman and Johansson, Fredrik and Matthieu Komorowski and Aldo Faisal and Sontag, David and Finale Doshi-Velez and Celi, Leo} } @proceedings {630368, title = {An Evaluation of the Human-Interpretability of Explanation}, journal = {Conference on Neural Information Processing Systems (NeurIPS) Workshop on Correcting and Critiquing Trends in Machine Learning}, year = {2018}, author = {I. Lage and E. Chen and J. He and M. Narayanan and S. Gershman and B. Kim and Doshi-Velez, F.} } @proceedings {630276, title = {Projected BNNs: Avoiding weight-space pathologies by projecting neural network weights}, journal = { Conference on Neural Information Processing Systems (NeurIPS) Workshop on Bayesian Deep Learning }, year = {2018}, author = {Melanie F. Pradier and Weiwei Pan and Jiayu Yao and Soumya Ghosh and Finale Doshi-Velez} } @proceedings {630230, title = {Hierarchical Stick-breaking Feature Paintbox}, journal = {Conference on Neural Information Processing Systems (NeurIPS) Workshop on All of Bayesian Nonparametrics}, year = {2018}, author = {M. Fernandez-Pradier and Pan, W. and M. Yao and R. Singh and Finale Doshi-Velez} } @proceedings {630045, title = {Prediction-Constrained POMDPs}, journal = {Conference on Neural Information Processing Systems (NeurIPS) Workshop on Reinforcement Learning under Partial Observability }, year = {2018}, author = {Joseph Futoma and Michael C. Hughes and Finale Doshi-Velez} } @article {630044, title = {Improving counterfactual reasoning with kernelised dynamic mixing models}, journal = {PLoS ONE }, volume = {13}, number = {11}, year = {2018}, author = {Parbhoo, Sonali and Omer Gottesman and Andrew Slavin Ross and Matthieu Komorowski and Aldo Faisal and Bon, Isabella and Roth, Volker and Finale Doshi-Velez} } @proceedings {630017, title = {Human-in-the-Loop Interpretability Prior}, journal = {Conference on Neural Information Processing Systems (NeurIPS)}, year = {2018}, author = {Isaac Lage and Andrew Ross and Kim, Been and Gershman, Samuel and Finale Doshi-Velez} } @proceedings {622764, title = {Beyond Sparsity: Tree Regularization of Deep Models for Interpretability}, journal = {Association for the Advancement of Artificial Intelligence (AAAI)}, year = {2018}, author = {Mike Wu and Michael Hughes and Parbhoo, Sonali and Zazzi, Maurizio and Roth, Volker and Finale Doshi-Velez} } @proceedings {622763, title = {Diversity-Inducing Policy Gradient: Using MMD to find a set of policies that are diverse in terms of stete-visitation}, journal = {International Conference on Machine Learning (ICML) Exploration in Reinforcement Learning Workshop}, year = {2018}, author = {Muhammad A Masood and Finale Doshi-Velez} } @proceedings {622404, title = {Improving Sepsis Treatment Strategies by Combining Deep and Kernel-Based Reinforcement Learning}, journal = {American Medical Informatics Association (AMIA) Annual Symposium}, year = {2018}, author = {Xuefeng Peng and Yi Ding and David Wihl and Omer Gottesman and Matthieu Komorowski and Li-wei H. Lehman and Andrew Ross and Aldo Faisal and Finale Doshi-Velez} } @conference {622402, title = {Structured Variational Learning of Bayesian Neural Networks with Horseshoe Priors}, booktitle = {Proceedings of the 35th International Conference on Machine Learning (ICML)}, volume = {80}, year = {2018}, address = {Stockholm, Sweden}, author = {Soumya Ghosh and Jiayu Yao and Finale Doshi-Velez} } @conference {622400, title = {Decomposition of Uncertainty in Bayesian Deep Learning for Efficient and Risk-sensitive Learning}, booktitle = {Proceedings of the 35th International Conference on Machine Learning (ICML)}, volume = {80}, year = {2018}, address = {Stockholm, Sweden}, author = {Depeweg, Stefan and Jose Miguel Hernandez-Lobato and Finale Doshi-Velez and Steffen Udluft} } @conference {622398, title = {Weighted Tensor Decomposition for Learning Latent Variables with Partial Data}, booktitle = {Proceedings of the 21st International Conference on Artificial Intelligence and Statistics (AISTATS) 2018}, volume = {84}, year = {2018}, address = {Lanzarote, Spain}, author = {Omer Gottesman and Weiewei Pan and Finale Doshi-Velez} } @conference {622397, title = {Semi-Supervised Prediction-Constrained Topic Models}, booktitle = {Proceedings of the 21st International Conference on Artificial Intelligence and Statistics (AISTATS) 2018}, volume = {84}, year = {2018}, address = {Lanzarote, Spain}, author = {Michael C. Hughes and Gabriel Hope and Leah Weiner and Thomas H. McCoy, Jr. and Roy H. Perlis and Erik Sudderth and Finale Doshi-Velez} } @proceedings {622396, title = {Agent Strategy Summarization}, journal = {Autonomous Agents and Multiagent Systems, Blue Sky Ideas Track}, year = {2018}, author = {Amir, Ofra and Finale Doshi-Velez and David Sarne} } @proceedings {622395, title = {Accountability of AI Under the Law: The Role of Explanation}, journal = {Privacy Law Scholars Conference}, year = {2018}, author = {Finale Doshi-Velez and Mason Kortz and Ryan Budish and Chris Bavitz and Sam Gershman and David O{\textquoteright}Brien and Shieber, Stuart and James Waldo and David Weinberger and Alexandra Wood} } @proceedings {622393, title = {Unsupervised Grammar Induction with Depth-bounded PCFG}, journal = {Association for Computational Linguistics}, year = {2018}, author = {Lifeng Jin and Finale Doshi-Velez and Miller, Timothy and Schuler, William and Schwartz, Lane} } @proceedings {622392, title = {Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients}, journal = {Association for the Advancement of Artificial Intelligence (AAAI)}, year = {2018}, author = {Andrew Slavin Ross and Finale Doshi-Velez} } @proceedings {622264, title = {Direct Policy Transfer via Hidden Parameter Markov Decision Processes}, journal = {International Conference on Machine Learning (ICML) Workshop on Lifelong Learning,}, year = {2018}, author = {Jiayu Yao and Taylor Killian and George Konidaris and Finale Doshi-Velez} } @proceedings {622263, title = {Behaviour Policy Estimation in Off-Policy Policy Evaluation: Calibration Matters}, journal = {International Conference on Machine Learning (ICML) Workshop on CausalML}, year = {2018}, author = {Aniruddh Raghu and Omer Gottesman and Yao Liu and Matthieu Komorowski and Aldo Faisal and Finale Doshi-Velez and Emma Brunskill} } @proceedings {622261, title = {Stitched Trajectories for Off-Policy Learning}, journal = {International Conference on Machine Learning (ICML) Workshop on CausalML,}, year = {2018}, author = {Scott Sussex and Omer Gottesman and Yao Liu and Susan Murphy and Emma Brunskill and Finale Doshi-Velez} } @proceedings {622259, title = {Representation Balancing MDPs for Off-Policy Policy Evaluation}, journal = {International Conference on Machine Learning (ICML) Workshop on CausalML}, year = {2018}, author = {Yao Liu and Omer Gottesman and Aniruddh Raghu and Matthieu Komorowski and Aldo Faisal and Finale Doshi-Velez and Emma Brunskill} } @proceedings {622257, title = {Regularizing Tensor Decomposition Methods by Optimizing Pseudo-Data}, journal = {International Conference on Machine Learning (ICML) Exploration in Reinforcement Learning Workshop,}, year = {2018}, author = {Omer Gottesman and Finale Doshi-Velez} } @proceedings {622255, title = {Learning Qualitatively Diverse and Interpretable Rules for Classification}, journal = {International Conference on Machine Learning (ICML) Workshop on Human Interpretability in Machine Learning,}, year = {2018}, author = {Andrew Slavin Ross and Weiwei Pan and Finale Doshi-Velez} } @proceedings {621782, title = {Depth-bounding is effective: Improvements and Evaluation of Unsupervised PCFG Induction}, journal = {Conference on Empirical Methods in Natural Language Processing (EMNLP) }, year = {2018}, author = {Lifeng Jin and Finale Doshi-Velez and Miller, Tim and Schuler, William and Schwartz, Lane} } @inbook {621779, title = {Considerations for Evaluation and Generalization in Interpretable Machine Learning}, booktitle = {Explainable and Interpretable Models in Computer Vision and Machine Learning}, year = {2018}, publisher = {Springer International Publishing}, organization = {Springer International Publishing}, edition = {1st}, author = {Finale Doshi-Velez and Kim, Been}, editor = {Hugo Escalante and Sergio Escalera and Isabelle Guyon and Xavier Bar{\'o} and Ya{\u g}mur G{\"u}{\c c}l{\"u}t{\"u}rk and Umut G{\"u}{\c c}l{\"u} and Marcel A. J. van Gerven} } @article {604845, title = {PhenoLines: Phenotype Comparison Visualizations for Disease Subtyping via Topic Models}, journal = {IEEE Transactions on Visualization and Computer Graphics}, volume = {24}, number = {1}, year = {2018}, pages = {371-381}, author = {Michael Glueck and Mahdi Pakdaman Naeini and Finale Doshi-Velez and Fanny Chevalier and Azam Khan and Daniel Wigdor and Brudno, Michael} } @proceedings {622391, title = {Predicting intervention onset in the ICU with switching state space models}, journal = {American Medical Informatics Association (AMIA),}, year = {2017}, author = {Marzyeh Ghassemi and Mike Wu and Michael C. Hughes and Szolovits, Peter and Finale Doshi-Velez} } @proceedings {622360, title = {PhenoLines: Phenotype Comparison Visualizations for Disease Subtyping via Topic Models}, journal = {Conference on Visual Analytics Science and Technology (VAST),}, year = {2017}, author = {Michael Glueck and Mahdi Pakdaman Naeini and Finale Doshi-Velez and Fanny Chevalier and Azam Khan and Daniel Wigdor and Brudno, Michael} } @proceedings {622251, title = {The Neural LASSO: Local Linear Sparsity for Interpretable Explanations}, journal = {Neural Information Processing Systems (NIPS) Workshop on Transparent and Interpretable Machine Learning in Safety Critical Environments}, year = {2017}, author = {Andrew Slavin Ross and Isaac Lage and Finale Doshi-Velez} } @proceedings {622239, title = {Beyond Sparsity: Tree Regularization of Deep Models for Interpretability}, journal = {Neural Information Processing Systems (NIPS) Workshop on Transparent and Interpretable Machine Learning in Safety Critical Environments}, year = {2017}, author = {Mike Wu and Michael Hughes and Parbhoo, Sonali and Zazzi, Maurizio and Roth, Volker and Finale Doshi-Velez} } @proceedings {622237, title = {Counterfactual Reasoning with Dynamic Switching Models for HIV Therapy Selection}, journal = {Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Healthcare}, year = {2017}, author = {S. Parbhoo and V. Roth and Finale Doshi-Velez} } @proceedings {622236, title = {Prediction-Constrained Topic Models for Antidepressant Recommendation}, journal = {Neural Information Processing Systems (NIPS) Workshop on Machine Learning for Healthcare}, year = {2017}, author = {Michael C. Hughes and Gabriel Hope and Leah Weiner and Thomas H. McCoy and Roy H. Perlis and Erik B. Sudderth and Finale Doshi-Velez} } @proceedings {622235, title = {Model Selection in Bayesian Neural Networks via Horseshoe Priors}, journal = {Neural Information Processing Systems (NIPS) Workshop on Bayesian Deep Learning}, year = {2017}, author = {Soumya Ghosh and Finale Doshi-Velez} } @proceedings {622234, title = {Structured Variational Autoencoders for the Beta-Bernoulli Process}, journal = {Neural Information Processing Systems (NIPS) Workshop on Advances in Approximate Bayesian Inference}, year = {2017}, author = {Rachit Singh and Jeffrey Ling and Finale Doshi-Velez} } @proceedings {622233, title = {Clustering LaTeX Solutions to Machine Learning Assignments for Rapid Assessment}, journal = {Advancing Education with Data Knowledge Discovery and Data Mining (KDD) Workshop}, year = {2017}, author = {Sindy Tan and Finale Doshi-Velez and Juan Quiroz and Elena Glassman} } @proceedings {622232, title = {Uncertainty Decomposition in Bayesian Neural Networks with Latent Variables}, journal = {International Conference on Machine Learning (ICML) Workshop}, year = {2017}, author = {Depeweg, Stefan and Jose Miguel Hernandez-Lobato and Finale Doshi-Velez and Steffen Udluft} } @conference {621797, title = {Combining Kernel and Model Based Learning for HIV Therapy Selection}, booktitle = {AMIA Summits on Translational Science Proceedings }, volume = {2017}, year = {2017}, pages = {239}, author = {Parbhoo, Sonali and Bogojeska, Jasmina and Zazzi, Maurizio and Roth, Volker and Finale Doshi-Velez} } @proceedings {621792, title = {Prior Matters: Simple and General Methods for Evaluating and Improving Topic Quality in Topic Modeling}, journal = {Text as Data}, year = {2017}, author = {Angela Fan and Finale Doshi-Velez and Miratrix, Luke} } @proceedings {605255, title = {Robust and Efficient Transfer Learning with Hidden Parameter Markov Decision Processes}, journal = {Neural Information Processing Systems (NIPS)}, year = {2017}, address = {Long Beach, CA}, author = {Taylor Killian and Samuel Daulton and George Konidaris and Finale Doshi-Velez} } @article {604846, title = {A Bayesian Framework for Learning Rule Sets for Interpretable Classification}, journal = {Journal of Machine Learning}, volume = {18}, number = {70}, year = {2017}, pages = {1-37}, author = {Wang, Tong and Rudin, Cynthia and Finale Doshi-Velez and Liu, Yimin and Klampfl, Erica and MacNeille, Perry} } @proceedings {588961, title = {Right for the Right Reasons: Training Differentiable Models by Constraining their Explananations}, journal = {International Joint Conference on Artificial Intelligence (IJCAI)}, year = {2017}, address = {Melbourne, Australia}, author = {Andrew Slavin Ross and Michael C. Hughes and Finale Doshi-Velez} } @article {557146, title = {Restricted Indian Buffet Processes}, journal = {Statistics and Computing}, volume = {27}, number = {5}, year = {2017}, pages = {1205-1223}, author = {Finale Doshi-Velez and Sinead Williamson} } @article {463811, title = {Understanding Vasopressor Intervention and Weaning: Risk Prediction in a Public Heterogeneous Clinical Time Series Database}, journal = {Journal of the American Medical Informatics Association}, volume = {24}, number = {3}, year = {2017}, pages = {488-495}, abstract = { \  Background The widespread adoption of electronic health records allows us to ask evidence-based questions about the need for and benefits of specific clinical interventions in critical-care settings across large populations. Objective We investigated the prediction of vasopressor administration and weaning in the intensive care unit. Vasopressors are commonly used to control hypotension, and changes in timing and dosage can have a large impact on patient outcomes. Materials and Methods We considered a cohort of 15 695 intensive care unit patients without orders for reduced care who were alive 30 days post-discharge. A switching-state autoregressive model (SSAM) was trained to predict the multidimensional physiological time series of patients before, during, and after vasopressor administration. The latent states from the SSAM were used as predictors of vasopressor administration and weaning. Results The unsupervised SSAM features were able to predict patient vasopressor administration and successful patient weaning. Features derived from the SSAM achieved areas under the receiver operating curve of 0.92, 0.88, and 0.71 for predicting ungapped vasopressor administration, gapped vasopressor administration, and vasopressor weaning, respectively. We also demonstrated many cases where our model predicted weaning well in advance of a successful wean. Conclusion Models that used SSAM features increased performance on both predictive tasks. These improvements may reflect an underlying, and ultimately predictive, latent state detectable from the physiological time series. \  }, author = {Mike Wu and Marzyeh Ghassemi and Mengling Fend and Leo A. Celi and Szolovits, Peter and Finale Doshi-Velez} } @article {453946, title = {Learning and Policy Search in Stochastic Dynamical Systems with Bayesian Neural Networks}, journal = {ICLR}, year = {2017}, abstract = { We present an algorithm for model-based reinforcement learning that combines Bayesian neural networks (BNNs) with random roll-outs and stochastic optimization for policy learning. The BNNs are trained by minimizing α\  -divergences, allowing us to capture complicated statistical patterns in the transition dynamics, e.g. multi-modality and heteroskedasticity, which are usually missed by other common modeling approaches. We illustrate the performance of our method by solving a challenging benchmark where model-based approaches usually fail and by obtaining promising results in a real-world scenario for controlling a gas turbine. }, author = {Stefan Depewag and Jos{\'e} Miguel Hern{\'a}ndez-Lobato and Finale Doshi-Velez and Steffen Udluft} } @proceedings {622762, title = {Combining Kernel and Model Based Learning for HIV Therapy Selection}, journal = {Neural Information Processing Systems (NIPS) Workshop for Machine Learning and Healthcare}, year = {2016}, author = {Parbhoo, Sonali and Bogojeska, Jasmina and Zazzi, Maurizio and Roth, Volker and Finale Doshi-Velez} } @proceedings {622231, title = {Transfer Learning Across Patient Variations with Hidden Parameter Markov Decision Processes}, journal = {Neural Information Processing Systems (NIPS) Workshop for Machine Learning and Healthcare}, year = {2016}, author = {Taylor W. Killian and George Konidaris and Finale Doshi-Velez} } @proceedings {622230, title = {Supervised topic models for clinical interpretability}, journal = {Neural Information Processing Systems (NIPS) Workshop for Machine Learning and Healthcare}, year = {2016}, author = {Michael C. Hughes and Elibol, Huseyin Melih and Thomas McCoy and Roy Perlis and Finale Doshi-Velez} } @proceedings {622194, title = {Robust Posterior Exploration in NMF}, journal = {International Conference on Machine Learning (ICML) Workshop on Geometry in Machine Learning}, year = {2016}, author = {Muhammad Arjumand Masood and Finale Doshi-Velez} } @proceedings {588966, title = {Memory-Bounded Left-Corner Unsupervised Grammar Induction on Child-Directed Input}, journal = {Computational Linguistics: Technical Papers (COLING)}, year = {2016}, pages = {964-975}, publisher = {COLING}, address = {Osaka, Japan}, author = {Cory Shain and William Bryce and Lifeng Jin and Victoria Krakovna and Finale Doshi-Velez and Miller, Timothy and Schuler, William and Schwartz, Lane} } @proceedings {552811, title = {Hidden Parameter Markov Decision Processes: A Semiparametric Regression Approach for Discovering Latent Task Parametrizations}, journal = {IJCAI}, year = {2016}, author = {Doshi-Velez, F. and Konidaris, G} } @article {464471, title = {Cross-Corpora Unsupervised Learning of Trajectories in Autism Spectrum Disorders}, journal = {Journal of Machine Learning Research}, volume = {17}, number = {1}, year = {2016}, pages = {4597-4634}, author = {Melih Elibol and Vincent Nguyen and Scott Linderman and Matthew Johnson and Amna Hashmi and Finale Doshi-Velez} } @article {464541, title = {Spectral M-estimation with Application to Hidden Markov Models: Supplementary Material}, journal = {AISTATS}, year = {2016}, author = {Dustin Tran and Minjae Kim and Finale Doshi-Velz} } @article {453961, title = {A Characterization of the Non-Uniqueness of Nonnegative Matrix Factorizations}, journal = {arXiv:1604.00653 }, year = {2016}, abstract = {Nonnegative matrix factorization (NMF) is a popular dimension reduction technique that produces interpretable decomposition of the data into parts. However, this decompostion is not generally identifiable (even up to permutation and scaling). While other studies have provide criteria under which NMF is identifiable, we present the first (to our knowledge) characterization of the non-identifiability of NMF. We describe exactly when and how non-uniqueness can occur, which has important implications for algorithms to efficiently discover alternate solutions, if they exist.}, author = {Weiwei Pan and Finale Doshi-Velez} } @article {453941, title = {Cost-Sensitive Batch Mode Active Learning: Designing Astronomical Observation by Optimizing Telescope Time and Telescope Choice}, year = {2016}, abstract = { }, author = {Xide Xia and Pavlos Protopapas and Finale Doshi-Velez} } @article {453956, title = {An Empirical Comparison of Sampling Quality Metrics: A Case Study for Bayesian Nonnegative Matrix Factorization}, journal = {arXiv preprint arXiv:1606.06250}, year = {2016}, abstract = {In this work, we empirically explore the question: how can we assess the quality of samples from some target distribution? We assume that the samples are provided by some valid Monte Carlo procedure, so we are guaranteed that the collection of samples will asymptotically approximate the true distribution. Most current evaluation approaches focus on two questions: (1) Has the chain mixed, that is, is it sampling from the distribution? and (2) How independent are the samples (as MCMC procedures produce correlated samples)? Focusing on the case of Bayesian nonnegative matrix factorization, we empirically evaluate standard metrics of sampler quality as well as propose new metrics to capture aspects that these measures fail to expose. The aspect of sampling that is of particular interest to us is the ability (or inability) of sampling methods to move between multiple optima in NMF problems. As a proxy, we propose and study a number of metrics that might quantify the diversity of a set of NMF factorizations obtained by a sampler through quantifying the coverage of the posterior distribution. We compare the performance of a number of standard sampling methods for NMF in terms of these new metrics.}, author = {Arjumand Masood and Weiwei Pan and Finale Doshi-Velez} } @article {453966, title = {Machine Learning Approaches to Environmental Disturbance Rejection in Multi-Axis Optoelectronic Force Sensors}, journal = {Sensors and Actuators A: Physical}, volume = {248}, year = {2016}, pages = {78-87}, abstract = { Light-intensity modulated (LIM) force sensors are seeing increasing interest in the field of surgical robotics and flexible systems in particular. However, such sensing modalities are notoriously susceptible to ambient effects such as temperature and environmental irradiance which can register as false force readings. We explore machine learning techniques to dynamically compensate for environmental biases that plague multi-axis optoelectronic force sensors. In this work, we fabricate a multisensor: three-axis LIM force sensor with integrated temperature and ambient irradiance sensing manufactured via a monolithic, origami-inspired fabrication process called printed-circuit MEMS. We explore machine learning regression techniques to compensate for temperature and ambient light sensitivity using on-board environmental sensor data. We compare batch-based ridge regression, kernelized regression and support vector techniques to baseline ordinary least-squares estimates to show that on-board environmental monitoring can substantially improve sensor force tracking performance and output stability under variable lighting and large (\>100\ {\textdegree}C) thermal gradients. By augmenting the least-squares estimate with nonlinear functions describing both environmental disturbances and cross-axis coupling effects, we can reduce the error in Fx, Fy and Fz by 10\%, 33\%, and 73\%, respectively. We assess viability of each algorithm tested in terms of both prediction accuracy and computational overhead, and analyze kernel-based regression for prediction in the context of online force feedback and haptics applications in surgical robotics. Finally, we suggest future work for fast approximation and prediction using stochastic, sparse kernel techniques. }, author = {Joshua Gafford and Finale Doshi-Velez and Robert Wood and Walsh, Conor} } @article {452686, title = {Electronic Health Record Based Algorithm to Identify Patients with Autism Spectrum Disorder}, journal = {PLoS ONE 11(7): e0159621}, year = {2016}, author = {Todd Lingren and Pei Chen and Joseph Bochenek and Finale Doshi-Velez and Patty Manning-Courtney and Julie Bickel and Leah Wildenger Welchons and Judy Reinhold and Nicole Bing and Yizhao Ni and William Barbaresi and Frank Mentch and Melissa Basford and Joshua Denny and Lyam Vazquez and Cassandra Perry and Bahram Namjou and Haijun Qiu and John Connolly and Debra Abrams and Ingrid A. Holm and Beth A. Cobb and Nataline Lingren and Imre Solti and Hakonarson, Hakon and Isaac S. Kohane and Harley, John and Savova, Guergana} } @article {452676, title = {Increasing the Interpretability of Recurrent Neural Networks Using Hidden Markov Models}, journal = {arXiv:1606.05320 }, year = {2016}, abstract = {Abstract: As deep neural networks continue to revolutionize various application domains, there is increasing interest in making these powerful models more understandable and interpretable, and narrowing down the causes of good and bad predictions. We focus on recurrent neural networks (RNNs), state of the art models in speech recognition and translation. Our approach to increasing interpretability is by combining an RNN with a hidden Markov model (HMM), a simpler and more transparent model. We explore various combinations of RNNs and HMMs: an HMM trained on LSTM states; a hybrid model where an HMM is trained first, then a small LSTM is given HMM state distributions and trained to fill in gaps in the HMM{\textquoteright}s performance; and a jointly trained hybrid model. We find that the LSTM and HMM learn complementary information about the features in the text.}, author = {Viktoriya Krakovna and Finale Doshi-Velez} } @article {wang2015or, title = {Bayesian Or{\textquoteright}s of And{\textquoteright}s for Interpretable Classification with Application to Context Aware Recommender Systems}, journal = {arXiv:1504.07614}, year = {2015}, author = {Wang, Tong and Rudin, Cynthia and Finale Doshi-Velez and Liu, Yimin and Klampfl, Erica and MacNeille, Perry} } @conference {kim2015mind, title = {Mind the Gap: A Generative Approach to Interpretable Feature Selection and Extraction}, booktitle = {Advances in Neural Information Processing Systems}, year = {2015}, pages = {2251{\textendash}2259}, author = {Kim, Been and Shah, Julie A and Finale Doshi-Velez} } @article {doshi2015prevalence, title = {Prevalence of Inflammatory Bowel Disease Among Patients with Autism Spectrum Disorders}, journal = {Inflammatory bowel diseases}, volume = {21}, year = {2015}, pages = {2281{\textendash}2288}, publisher = {LWW}, author = {Finale Doshi-Velez and Avillach, Paul and Palmer, Nathan and Bousvaros, Athos and Yaorong Ge and Fox, Kathe and Steinberg, Greg and Spettell, Claire and Juster, Iver and Isaac Kohane} } @article {247726, title = {Bayesian Nonparametric Methods for Partially-Observable Reinforcement Learning}, journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence }, volume = {37}, number = {2}, year = {2015}, pages = {394 - 407}, abstract = { Making intelligent decisions from incomplete information is critical in many applications: for example, robots must choose actions based on imperfect sensors, and speech-based interfaces must infer a user{\textquoteright}s needs from noisy microphone inputs. What makes these tasks hard is that often we do not have a natural representation with which to model the domain and use for choosing actions; we must learn about the domain{\textquoteright}s properties while simultaneously performing the task. Learning a representation also involves trade-offs between modeling the data that we have seen previously and being able to make predictions about new data. This article explores learning representations of stochastic systems using Bayesian nonparametric statistics. Bayesian nonparametric methods allow the sophistication of a representation to scale gracefully with the complexity in the data. Our main contribution is a careful empirical evaluation of how representations learned using Bayesian nonparametric methods compare to other standard learning approaches, especially in support of planning and control. We show that the Bayesian aspects of the methods result in achieving state-of-the-art performance in decision making with relatively few samples, while the nonparametric aspects often result in fewer computations. These results hold across a variety of different techniques for choosing actions given a representation. }, author = {Finale Doshi-Velez and David Pfau and Frank Wood and Nicholas Roy} } @article {247731, title = {HackEbola with Data: On the hackathon format for timely data analysis.}, year = {2015}, abstract = {For more information, see the event page:\ https://projects.iq.harvard.edu/hack}, author = {Finale Doshi-Velez and Yael E. Marshall} } @article {246626, title = {Graph-Sparse LDA: A Topic Model with Structured Sparsity}, journal = {AAAI }, year = {2015}, abstract = {Originally designed to model text, topic modeling has become a powerful tool for uncovering latent structure in domains including medicine, finance, and vision. The goals for the model vary depending on the application: in some cases, the discovered topics may be used for prediction or some other downstream task. In other cases, the content of the topic itself may be of intrinsic scientific interest. Unfortunately, even using modern sparse techniques, the discovered topics are often difficult to interpret due to the high dimensionality of the underlying space. To improve topic interpretability, we introduce Graph-Sparse LDA, a hierarchical topic model that leverages knowledge of relationships between words (e.g., as encoded by an ontology). In our model, topics are summarized by a few latent concept-words from the underlying graph that explain the observed words. Graph-Sparse LDA recovers sparse, interpretable summaries on two real-world biomedical datasets while matching state-of-the-art prediction performance.}, author = {Finale Doshi-Velez and Byron C. Wallace and Adams, Ryan P} } @article {doshi2014comorbidity, title = {Comorbidity clusters in autism spectrum disorders: an electronic health record time-series analysis}, journal = {Pediatrics}, volume = {133}, year = {2014}, pages = {e54{\textendash}e63}, publisher = {American Academy of Pediatrics}, author = {Finale Doshi-Velez and Yaorong Ge and Isaac Kohane} } @article {doshi2014graph, title = {Graph-Sparse LDA: A Topic Model with Structured Sparsity}, journal = {arXiv:1410.4510}, year = {2014}, author = {Finale Doshi-Velez and Wallace, Byron and Adams, Ryan} } @proceedings {246631, title = {Hidden Parameter Markov Decision Processes: An Emerging Paradigm for Modeling Families of Related Tasks}, journal = {AAAI 2014 Fall Symposium on Knowledge, Skill, and Behavior Transfer in Autonomous Robots}, year = {2014}, abstract = {The goal of transfer is to use knowledge obtained by solving one task to improve a robot{\textquoteright}s (or software agent{\textquoteright}s) performance in future tasks. In general, we do not expect this to work; for transfer to be feasible, there must be something in common between the source task(s) and goal task(s). The question at the core of the transfer learning enterprise is therefore: what makes two tasks related?, or more generally, how do you define a family of related tasks? Given a precise definition of how a particular family of tasks is related, we can formulate clear optimization methods for selecting source tasks and determining what knowledge should be imported from the source task(s), and how it should be used in the target task(s). This paper describes one model that has appeared in several different research scenarios where an agent is faced with a family of tasks that have similar, but not identical, dynamics (or reward functions). For example, a human learning to play baseball may, over the course of their career, be exposed to several different bats, each with slightly different weights and lengths. A human who has learned to play baseball well with one bat would be expected to be able to pick up any similar bat and use it. Similarly, when learning to drive a car, one may learn in more than one car, and then be expected to be able to drive any make and model of car (within reasonable variations) with little or no relearning. These examples are instances of exactly the kind of flexible, reliable, and sample-efficient behavior that we should be aiming to achieve in robotics applications. One way to model such a family of tasks is to posit that they are generated by a small set of latent parameters (e.g., the length and weight of the bat, or parameters describing the various physical properties of the car{\textquoteright}s steering system and clutch) that are fixed for each problem instance (e.g., for each bat, or car), but are not directly observable by the agent. Defining a distribution over these latent parameters results in a family of related tasks, and transfer is feasible to the extent that the number of latent variables is small, the task dynamics (or reward function) vary smoothly with them, and to the extent to which they can either be ignored or identified using transition data from the task. This model has appeared\ under several different names in the literature; we refer to it as a hidden-parameter Markov decision process (or HIPMDP).}, author = {George Konidaris and Finale Doshi-Velez} } @proceedings {246611, title = {Unfolding Physiological State: Mortality Modelling in Intensive Care Units}, journal = {ACM SIGKDD international conference on Knowledge discovery and data mining}, year = {2014}, pages = {75-84 }, address = {New York City}, abstract = {Accurate knowledge of a patient{\textquoteright}s disease state and trajectory is critical in a clinical setting. Modern electronic healthcare records contain an increasingly large amount of data, and the ability to automatically identify the factors that influence patient outcomes stand to greatly improve the ef- ficiency and quality of care. We examined the use of latent variable models (viz. Latent Dirichlet Allocation) to decompose free-text hospital notes into meaningful features, and the predictive power of these features for patient mortality. We considered three prediction regimes: (1) baseline prediction, (2) dynamic (timevarying) outcome prediction, and (3) retrospective outcome prediction. In each, our prediction task differs from the familiar time-varying situation whereby data accumulates; since fewer patients have long ICU stays, as we move forward in time fewer patients are available and the prediction task becomes increasingly difficult. We found that latent topic-derived features were effective in determining patient mortality under three timelines: inhospital, 30 day post-discharge, and 1 year post-discharge mortality. Our results demonstrated that the latent topic features important in predicting hospital mortality are very different from those that are important in post-discharge. mortality. In general, latent topic features were more predictive than structured features, and a combination of the two performed best. The time-varying models that combined latent topic features and baseline features had AUCs that reached 0.85, 0.80, and 0.77 for in-hospital, 30 day post-discharge and 1 year post-discharge mortality respectively. Our results agreed with other work suggesting that the first 24 hours of patient information are often the most predictive of hospital mortality. Retrospective models that used a combination of latent topic features and structured features achieved AUCs of 0.96, 0.82, and 0.81 for in-hospital, 30 day, and 1-year mortality prediction. Our work focuses on the dynamic (time-varying) setting, because models from this regime could facilitate an on-going severity stratification system that helps d}, author = {Marzyeh Ghassemi and Tristan Naumann and Finale Doshi-Velez and Nicole Brimmer and Rohit Joshi and Anna Rumshisky and Szolovits, Peter} } @article {246606, title = {Comorbidity Clusters in Autism Spectrum Disorders: An Electronic Health Record Time-Series Analysis}, journal = {Pediatrics}, volume = {10.1542}, number = {peds.2013}, year = {2013}, pages = {0819}, abstract = { OBJECTIVE:\ The distinct trajectories of patients with autism spectrum disorders (ASDs) have not been extensively studied, particularly regarding clinical manifestations beyond the neurobehavioral criteria from the Diagnostic and Statistical Manual of Mental Disorders. The objective of this study was to investigate the patterns of co-occurrence of medical comorbidities in ASDs. METHODS:\ International Classification of Diseases, Ninth Revision\ codes from patients aged at least 15 years and a diagnosis of ASD were obtained from electronic medical records. These codes were aggregated by using phenotype-wide association studies categories and processed into 1350-dimensional vectors describing the counts of the most common categories in 6-month blocks between the ages of 0 to 15. Hierarchical clustering was used to identify subgroups with distinct courses. RESULTS:\ Four subgroups were identified. The first was characterized by seizures (n\ = 120, subgroup prevalence 77.5\%). The second (n\ = 197) was characterized by multisystem disorders including gastrointestinal disorders (prevalence 24.3\%) and auditory disorders and infections (prevalence 87.8\%), and the third was characterized by psychiatric disorders (n\ = 212, prevalence 33.0\%). The last group (n\ = 4316) could not be further resolved. The prevalence of psychiatric disorders was uncorrelated with seizure activity (P\ = .17), but a significant correlation existed between gastrointestinal disorders and seizures (P\ \< .001). The correlation results were replicated by using a second sample of 496 individuals from a different geographic region. CONCLUSIONS:\ Three distinct patterns of medical trajectories were identified by unsupervised clustering of electronic health record diagnoses. These may point to distinct etiologies with different genetic and environmental contributions. Additional clinical and molecular characterizations will be required to further delineate these subgroups. }, author = {Finale Doshi-Velez and Yaorong Ge and Isaac Kohane} } @article {DBLP:journals/corr/Doshi-VelezK13, title = {Hidden Parameter Markov Decision Processes: A Semiparametric Regression Approach for Discovering Latent Task Parametrizations}, journal = {CoRR}, volume = {abs/1308.3513}, year = {2013}, abstract = {Control applications often feature tasks with similar, but not identical, dynamics. We introduce the Hidden Parameter Markov Decision Process (HiP-MDP), a framework that parametrizes a family of related dynamical systems with a low-dimensional set of latent factors, and introduce a semiparametric regression approach for learning its structure from data. In the control setting, we show that a learned HiP-MDP rapidly identifies the dynamics of a new task instance, allowing an agent to flexibly adapt to task variations.}, author = {Finale Doshi-Velez and George Konidaris} } @mastersthesis {doshi2012bayesian, title = {Bayesian nonparametric approaches for reinforcement learning in partially observable domains}, year = {2012}, school = {Massachusetts Institute of Technology}, type = {phd}, author = {Finale Doshi-Velez} } @article {doshi2012improving, title = {Improving safety and operational efficiency in residential care settings with WiFi-based localization}, journal = {Journal of the American Medical Directors Association}, volume = {13}, year = {2012}, pages = {558{\textendash}563}, publisher = {Elsevier}, author = {Finale Doshi-Velez and William Li and Yoni Battat and Ben Charrow and Dorothy Curthis and Jun-Geun Park and Hemachandra, Sachithra and Javier Velez and Cynthia Walsh and Don Fredette and others} } @article {doshi2012transfer, title = {Transfer Learning by Discovering Latent Task Parametrizations}, journal = {the NIPS 2012 Workshop on Bayesian Nonparametric Models for Reliable Planning And Decision-Making Under Uncertainty}, year = {2012}, author = {Finale Doshi-Velez and George Konidaris} } @article {DBLP:conf/icra/JosephDR12, title = {A Bayesian nonparametric approach to modeling battery health}, journal = {IEEE International Conference on Robotics and Automation}, year = {2012}, pages = {1876{\textendash}1882}, abstract = {Abstract{\textemdash}Making intelligent decisions from incomplete information is critical in many applications: for example, robots must choose actions based on imperfect sensors, and speech-based interfaces must infer a user{\textquoteright}s needs from noisy microphone inputs. What makes these tasks hard is that often we do not have a natural representation with which to model the domain and use for choosing actions; we must learn about the domain{\textquoteright}s properties while simultaneously performing the task. Learning a representation also involves trade-offs between modeling the data that we have seen previously and being able to make predictions about new data. This article explores learning representations of stochastic systems using Bayesian nonparametric statistics. Bayesian nonparametric methods allow the sophistication of a representation to scale gracefully with the complexity in the data. Our main contribution is a careful empirical evaluation of how representations learned using Bayesian nonparametric methods compare to other standard learning approaches, especially in support of planning and control. We show that the Bayesian aspects of the methods result in achieving state-of-the-art performance in decision making with relatively few samples, while the nonparametric aspects often result in fewer computations. These results hold across a variety of different techniques for choosing actions given a representation. Index Terms{\textemdash}Artificial intelligence, machine learning, reinforcement learning, partially-observable Markov decision process, hierarchial Dirichlet process hidden Markov model.}, doi = {10.1109/ICRA.2012.6225178}, author = {Joshua Mason Joseph and Finale Doshi-Velez and Nicholas Roy} } @proceedings {622359, title = {A Comparison of Human and Agent Reinforcement Learning in Partially Observable Domains}, journal = {33rd Annual Meeting of the Cognitive Science Society (CogSci)}, year = {2011}, author = {Finale Doshi-Velez and Zoubin Ghahramani} } @proceedings {622193, title = {An Analysis of Activity Changes in MS Patients: A Case Study in the Use of Bayesian Nonparametrics}, journal = {Neural Information Processing Systems (NIPS) Workshop: Bayesian Nonparametrics, Hope or Hype?}, year = {2011}, author = {Finale Doshi-Velez and Nicholas Roy} } @article {DBLP:journals/arobots/JosephDHR11, title = {A Bayesian nonparametric approach to modeling motion patterns}, journal = {Auton. Robots}, volume = {31}, year = {2011}, pages = {383{\textendash}400}, doi = {10.1007/s10514-011-9248-x}, author = {Joshua Mason Joseph and Finale Doshi-Velez and Albert S. Huang and Nicholas Roy} } @conference {DBLP:conf/icml/DoshiWTR11, title = {Infinite Dynamic Bayesian Networks}, booktitle = {Proceedings of the 28th International Conference on Machine Learning}, year = {2011}, pages = {913{\textendash}920}, author = {Finale Doshi and David Wingate and Joshua B. Tenenbaum and Nicholas Roy} } @conference {DBLP:conf/icml/GeramifardDRRH11, title = {Online Discovery of Feature Dependencies}, booktitle = {Proceedings of the 28th International Conference on Machine Learning}, year = {2011}, pages = {881{\textendash}888}, author = {Alborz Geramifard and Finale Doshi and Josh Redding and Nicholas Roy and Jonathan P. How} } @conference {DBLP:conf/aaai/JosephDR10, title = {A Bayesian Nonparametric Approach to Modeling Mobility Patterns}, booktitle = {Proceedings of the Twenty-Fourth Conference on Artificial Intelligence}, year = {2010}, author = {Joshua Mason Joseph and Finale Doshi-Velez and Nicholas Roy} } @conference {DBLP:conf/aaai/Doshi-Velez10, title = {Nonparametric Bayesian Approaches for Reinforcement Learning in Partially Observable Domains}, booktitle = { Conference on Artificial Intelligence}, year = {2010}, author = {Finale Doshi-Velez} } @conference {DBLP:conf/nips/Doshi-VelezWRT10, title = {Nonparametric Bayesian Policy Priors for Reinforcement Learning}, booktitle = {Advances in Neural Information Processing Systems 23: 24th Annual Conference on Neural Information Processing Systems 2010. Proceedings of a meeting held 6-9 December 2010, Vancouver, British Columbia, Canada.}, year = {2010}, pages = {532{\textendash}540}, author = {Finale Doshi-Velez and David Wingate and Nicholas Roy and Joshua B. Tenenbaum} } @proceedings {622265, title = {The Infinite Partially Observable Markov Decision Process}, journal = {Advances in Neural Information Processing Systems (NIPS)}, year = {2009}, author = {Finale Doshi-Velez} } @conference {237116, title = {Accelerated Sampling for the Indian Buffet Process}, booktitle = {Proceedings of the 26th International Conference on Machine Learning}, year = {2009}, address = {Montreal, Canada}, abstract = {We often seek to identify co-occurring hidden features in a set of observations. The Indian Buffet Process (IBP) provides a nonparametric prior on the features present in each observation, but current inference techniques for the IBP often scale poorly. The collapsed Gibbs sampler for the IBP has a running time cubic in the number of observations, and the uncollapsed Gibbs sampler, while linear, is often slow to mix. We present a new linear-time collapsed Gibbs sampler for conjugate likelihood models and demonstrate its efficacy on large real-world datasets.}, author = {Finale Doshi-Velez and Zoubin Ghahramani} } @conference {DBLP:conf/uai/Doshi-VelezG09, title = {Correlated Non-Parametric Latent Feature Models}, booktitle = {{UAI} 2009, Proceedings of the Twenty-Fifth Conference on Uncertainty in Artificial Intelligence, Montreal, QC, Canada, June 18-21, 2009}, year = {2009}, pages = {143{\textendash}150}, author = {Finale Doshi-Velez and Zoubin Ghahramani} } @proceedings {DBLP:conf/nips/Doshi-VelezKMG09, title = {Large Scale Nonparametric Bayesian Inference: Data Parallelisation in the Indian Buffet Process}, journal = {Conference on Neural Information Processing Systems (NIPS)}, year = {2009}, author = {Finale Doshi-Velez and David A. Knowles and Shakir Mohamed and Zoubin Ghahramani} } @proceedings {DBLP:journals/jmlr/DoshiMGT09, title = {Variational Inference for the Indian Buffet Process}, journal = {Artificial Intelligence on Statistics (AISTATS) Best Paper Nominee}, year = {2009}, author = {Finale Doshi and Kurt Miller and Jurgen Van Gael and Yee Whye Teh} } @article {237106, title = {Spoken Language Interaction with Model Uncertainty: An Adaptive Human-Robot Interaction System}, journal = {Connection Science}, volume = {20}, number = {4}, year = {2008}, pages = {299-318}, abstract = { Spoken language is one of the most intuitive forms of interaction between humans and agents. Unfortunately, agents that interact with people using natural language often experience communication errors and do not correctly understand the user{\textquoteright}s intentions. Recent systems have successfully used probabilistic models of speech, language, and user behavior to generate robust dialog performance in the presence of noisy speech recognition and ambiguous language choices, but decisions made using these probabilistic models are still prone to errors due to the complexity of acquiring and maintaining a complete model of human language and behavior. In this paper, we describe a decision-theoretic model for human-robot interaction using natural language. Our algorithm is based on the Partially Observable Markov Decision Process (POMDP), which allows agents to choose actions that are robust not only to uncertainty from noisy or ambiguous speech recognition but also unknown user models. Like most dialog systems, a POMDP is defined by a large number of parameters that may be difficult to specify a priori from domain knowledge, and learning these parameters from the user may require an unacceptably long training period. We describe an extension to the POMDP model that allows the agent to acquire a linguistic model of the user online, including new vocabulary and word choice preferences. Our approach not only avoids a training period of constant questioning as the agent learns, but also allows the agent to actively query for additional information when its uncertainty suggests a high risk of mistakes. We demonstrate our approach both in simulation and on a natural language interaction system for a robotic wheelchair application. Keywords: dialog management, human-computer interface, adaptive systems, online learning, partially observable Markov decision processes }, author = {Finale Doshia and Nicholas Roy} } @proceedings {DBLP:conf/atal/DoshiR08, title = {The Permutable POMDP: Fast Solutions to POMDPs for Preference Elicitation}, journal = {Conference on Autonomous Agents and Multi-Agent Systems (AAMAS) Best Paper Nominee}, year = {2008}, doi = {10.1145/1402383.1402454}, author = {Finale Doshi and Nicholas Roy} } @proceedings {DBLP:conf/icml/DoshiPR08, title = {Reinforcement Learning with Limited Reinforcement: Using Bayes Risk for Active Learning in POMDPs}, journal = {International Conference on Machine Learning (ICML)}, year = {2008}, doi = {10.1145/1390156.1390189}, author = {Finale Doshi and Joelle Pineau and Nicholas Roy} } @conference {621801, title = {Learning User Models with Limited Reinforcement: An Adaptive Human-Robot Interaction System}, booktitle = {Symposium on Language and Robotics (LANGRO)}, year = {2007}, author = {Finale Doshi and Nicholas Roy} } @proceedings {371661, title = {Efficient Model Learning for Dialog Management}, journal = {Conference on Human Robot Interaction (HRI)}, year = {2007}, author = {Finale Doshi and Nicholas Roy} } @proceedings {DBLP:conf/iros/DoshiBSKRTR07, title = {Collision Detection in Legged Locomotion Using Supervised Learning}, journal = {Conference on Intelligent Robots and Systems (IROS)}, year = {2007}, address = {San Diego CA}, doi = {10.1109/IROS.2007.4399538}, author = {Finale Doshi and Emma Brunskill and Alexander C. Shkolnik and Thomas Kollar and Khashayar Rohanimanesh and Russ Tedrake and Nicholas Roy} }