Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Zoubin GhahramaniSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works234 from public data
- 3,100
- Do no harm: a roadmap for responsible machine learning for health care1,228
In this Perspective, the authors present a framework, context and guidelines for accelerating the translation of machine-learning-based interventions in health care.
- Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing Their Input Gradients774
It is demonstrated that regularizing input gradients makes them more naturally interpretable as rationales for model predictions, and also exhibits robustness to transferred adversarial examples generated to fool all of the other models.
- Right for the Right Reasons: Training Differentiable Models by Constraining their Explanations717
This work introduces a method for efficiently explaining and regularizing differentiable models by examining and selectively penalizing their input gradients, which provide a normal to the decision boundary.
- Accountability of AI Under the Law: The Role of Explanation510
Contrary to popular wisdom of AI systems as indecipherable black boxes, it is found that this level of explanation should often be technically feasible but may sometimes be practically onerous; in the future, AI systems can and should be held to a similar standard of explanation as humans currently are.
- Guidelines for reinforcement learning in healthcare493
New guidelines for reinforcement learning for decisions about patient treatment are provided that are hoped will accelerate the rate at which observational cohorts can inform healthcare practice in a safe, risk-conscious manner.
- 454
- Comorbidity Clusters in Autism Spectrum Disorders: An Electronic Health Record Time-Series Analysis417
Three distinct patterns of medical trajectories were identified by unsupervised clustering of electronic health record diagnoses that may point to distinct etiologies with different genetic and environmental contributions in ASDs.
- Ethical and regulatory challenges of large language models in medicine331
Ethical concerns stemming from the perspectives of users, developers, and regulators are highlighted, notably focusing on data privacy and rights of use, data provenance, intellectual property contamination, and broad applications and plasticity of LLMs.
- Beyond Sparsity: Tree Regularization of Deep Models for Interpretability302
This work explicitly regularizes deep models so human users might step through the process behind their predictions in little time, and trains deep time-series models so their class-probability predictions have high accuracy while being closely modeled by decision trees with few nodes.
- How machine-learning recommendations influence clinician treatment selections: the example of antidepressant selection295
It is found that interacting with ML recommendations did not significantly improve clinicians’ treatment selection accuracy, assessed as concordance with expert psychopharmacologist consensus, compared to baseline scenarios in which clinicians made treatment decisions independently.
- Unfolding physiological state: mortality modelling in intensive care units233
This work examined the use of latent variable models to decompose free-text hospital notes into meaningful features, and found that latent topic-derived features were effective in determining patient mortality under three timelines: in-hospital, 30 day post- Discharge, and 1 year post-discharge mortality.
- Unfolding physiological state233
This work examined the use of latent variable models to decompose free-text hospital notes into meaningful features, and found that latent topic-derived features were effective in determining patient mortality under three timelines: in-hospital, 30 day post- Discharge, and 1 year post-discharge mortality.
- Designing AI for Trust and Collaboration in Time-Constrained Medical Decisions: A Sociotechnical Lens202
It is demonstrated how current trends in explainable AI may be inappropriate for clinical environments and paths towards designing these tools for real-world medical systems are considered.
- Considerations for Evaluation and Generalization in Interpretable Machine Learning192
A definitions of interpretability are discussed and described and when interpretability is needed (and when it is not) and a taxonomy for rigorous evaluation is described, and recommendations for researchers are made.
- A Bayesian nonparametric approach to modeling motion patterns164
This work proposes modeling target motion patterns as a mixture of Gaussian processes with a Dirichlet process prior over mixture weights, which provides an adaptive representation for each individual motion pattern and automatically adjusts the complexity of the motion model based on the available data.
- Human Evaluation of Models Built for Interpretability161
This work conducts carefully controlled human-subject experiments in two domains across three tasks based on human-simulatability through which it identifies specific types of complexity that affect performance more heavily than others–trends that are consistent across tasks and domains.
- A Roadmap for a Rigorous Science of Interpretability.160
There is little consensus on what interpretability in machine learning is and how to evaluate it for benchmarking, and the objective of this review is to chart a path toward the definition and rigorous evaluation of interpretability.
- Gathering Strength, Gathering Storms: The One Hundred Year Study on Artificial Intelligence (AI100) 2021 Study Panel Report142
The report concludes that AI has made a major leap from the lab to people's lives in recent years, which increases the urgency to understand its potential negative effects, and asks a set of 14 questions probing critical areas of AI development.
- Model Selection in Bayesian Neural Networks via Horseshoe Priors134
This work applies a horseshoe prior over node pre-activations of a Bayesian neural network, which effectively turns off nodes that do not help explain the data, and prevents the BNN from under-fitting even when the number of nodes required is grossly over-estimated.
- 121
- Robust and Efficient Transfer Learning with Hidden Parameter Markov Decision Processes117
This work introduces a new formulation of the Hidden Parameter Markov Decision Process (HiP-MDP), a framework for modeling families of related tasks using low-dimensional latent embeddings and replaces the original Gaussian Process-based model with a Bayesian Neural Network, enabling more scalable inference.
- Mind the Gap: a generative approach to interpretable feature selection and extraction115
The Mind the Gap Model is presented, an approach for interpretable feature extraction and selection that allows for the model to both optimize parameters related to interpretability and to directly report a global set of distinguishable dimensions to assist with further data exploration and hypothesis generation.
- 113
- The Infinite Partially Observable Markov Decision Process112
An infinite POMDP (iPOMDP) model is defined that does not require knowledge of the size of the state space and assumes that the number of visited states will grow as the agent explores its world and only models visited states explicitly.
- Artificial intelligence education: An evidence-based medicine approach for consumers, translators, and developers111
This work outlines a core curriculum for AI education of future consumers, translators, and developers, emphasizing the links between AI and EBM, with suggestions for how teaching may be integrated into existing curricula.
- 98
- 82
- Electronic Health Record Based Algorithm to Identify Patients with Autism Spectrum Disorder80
In a large cross-institutional cohort, co-occurrence patterns of comorbidities in ASDs provide further hypothetical evidence for distinct courses in ASD and open avenues for other large-scale EHR studies and individualized treatment of ASD.
- Exploring Computational User Models for Agent Policy Summarization78
An imitation learning-based approach to policy summarization is introduced; a mismatch between the model used to extract a summary and the model used to reconstruct the policy results in worse reconstruction quality; and it is demonstrated that it is important to carefully consider user models in policy summarization.
- 70
- Summarizing agent strategies69
A conceptual framework for strategy summarization is proposed, which is envisioning as a collaborative process that involves both agents and people, and possible testbeds that could be used to evaluate progress in research on strategy summarization are suggested.
- Predicting intervention onset in the ICU with switching state space models.67
This work trains unsupervised switching state autoregressive models on vital signs from the public MIMIC-III database to capture patient movement between physiological states to provide a path towards evidence-driven administration of clinical interventions.
- Accelerated sampling for the Indian Buffet Process66
This work presents a new linear-time collapsed Gibbs sampler for conjugate likelihood models and demonstrates its efficacy on large real-world datasets.
- 64
- 63
- Leveraging Factored Action Spaces for Efficient Offline Reinforcement Learning in Healthcare63
It is demonstrated that incorporating factored action spaces into value-based RL can result in better-performing policies and help an agent make more accurate inferences within underexplored regions of the state-action space when applying RL to observational datasets.
- 63
- Diversity-Inducing Policy Gradient: Using Maximum Mean Discrepancy to Find a Set of Diverse Policies61
This work derives a gradient-based optimization technique that can be combined with existing policy gradient methods to now identify diverse collections of well-performing policies and demonstrates its approach on benchmarks and a healthcare task.
- 60
- Understanding vasopressor intervention and weaning: risk prediction in a public heterogeneous clinical time series database59
Models that used switching-state autoregressive model features increased performance on both predictive tasks and may reflect an underlying, and ultimately predictive, latent state detectable from the physiological time series.
- Graph-Sparse LDA: A Topic Model with Structured Sparsity58
Graph-Sparse LDA is introduced, a hierarchical topic model that uses knowledge of relationships between words (e.g., as encoded by an ontology) to improve topic interpretability and recovers sparse, interpretable summaries on two real-world biomedical datasets while matching state-of-the-art prediction performance.
- Evaluating the Interpretability of Generative Models by Interactive Reconstruction57
A task to quantify the human-interpretability of generative model representations, where users interactively modify representations to reconstruct target instances and finds performance on this task much more reliably differentiates entangled and disentangled models than baseline approaches.
- Prevalence of Inflammatory Bowel Disease Among Patients with Autism Spectrum Disorders56
Across each population with different kinds of ascertainment, there was a consistent and statistically significant increased prevalance of IBD in patients with ASD than their respective controls and nationally reported rates for pediatric IBD.
- 54
- Contrastive Explanations That Anticipate Human Misconceptions Can Improve Human Decision-Making Skills53
Results from a large-scale experiment demonstrate that contrastive explanations significantly enhance users’ independent decision-making skills compared to unilateral explanations, without sacrificing decision accuracy, and demonstrates that integrating human reasoning into AI design can promote human skill development.
- Accuracy-Time Tradeoffs in AI-Assisted Decision Making under Time Pressure53
It is suggested that AI assistances have different accuracy-time tradeoffs when people are under time pressure compared to no time pressure, and how to adapt AI assistances in this setting is explored.
- Designing Reinforcement Learning Algorithms for Digital Interventions: Pre-Implementation Guidelines53
The PCS framework is extended, 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, and it is shown how the framework was used to design an RL algorithm for Oralytics, a mobile health study aiming to improve users’ tooth-brushing behaviors through the personalized delivery of intervention messages.
- 52
- Author Correction: Do no harm: a roadmap for responsible machine learning for health care49
An amendment to this paper has been published and can be accessed via a link at the top of the paper.
- Bayesian Nonparametric Methods for Partially-Observable Reinforcement Learning49
This article explores learning representations of stochastic systems using Bayesian nonparametric statistics and shows that the Bayesian aspects of the methods result in achieving state-of-the-art performance in decision making with relatively few samples, while the non parametric aspects often result in fewer computations.
- 47
- Regional Tree Regularization for Interpretability in Deep Neural Networks45
This work proposes regional tree regularization – a method that encourages a deep model to be well-approximated by several separate decision trees specific to predefined regions of the input space, and shows its approach delivers simpler explanations than other regularization schemes without compromising accuracy.
- A Bayesian Nonparametric Approach to Modeling Mobility Patterns44
This work proposes modeling the mobility patterns of moving agents as a mixture of Gaussian processes with a Dirichlet process prior over mixture weights, which provides a flexible representation for each individual mobility pattern.
- 43
- Large Scale Nonparametric Bayesian Inference: Data Parallelisation in the Indian Buffet Process43
This paper focuses on parallelisation of inference in the Indian Buffet Process, which allows data points to have an unbounded number of sparse latent features and the first parallel inference scheme for IBP-based models scales to datasets orders of magnitude larger than have previously been possible.
- Truly Batch Apprenticeship Learning with Deep Successor Features42
Deep Successor Feature Networks (DSFN) that estimate feature expectations in an off-policy setting and a transition-regularized imitation network that produces a near-expert initial policy and an efficient feature representation are introduced.
- Evaluating Machine Learning Articles42
A users’ guide to reading clinical machine learning articles featuring a synopsis of selected concepts in modern machine learning, and elaborate step-by-step guidance for physicians seeking to evaluate this evidence with a critical eye.
- Nonparametric Bayesian Policy Priors for Reinforcement Learning42
This work considers reinforcement learning in partially observable domains where the agent can query an expert for demonstrations and introduces priors that bias the agent towards models with both simple representations and simple policies, resulting in improved policy and model learning.
- Predicting treatment dropout after antidepressant initiation37
Machine learning applied to coded electronic health records facilitates identification of individuals at high-risk for treatment dropout following change in antidepressant medication, and may assist primary care physicians and psychiatrists in the clinic to personalize antidepressant treatment on the basis not solely of efficacy, but of tolerability.
- Semi-Supervised Prediction-Constrained Topic Models37
Improved prediction quality is demonstrated compared to previous supervised topic models, achieving results competitive with high-dimensional logistic regression on text analysis and electronic health records tasks while simultaneously learning interpretable topics.
- Unsupervised Grammar Induction with Depth-bounded PCFG36
Results for this model on grammar acquisition from transcribed child-directed speech and newswire text exceed or are competitive with those of other models when evaluated on parse accuracy, something which has not been demonstrated by other acquisition models.
- 34
- 34
- Optimizing for Interpretability in Deep Neural Networks with Tree Regularization33
This work explicitly regularizes deep models so that they are well-approximated by processes that humans can step through in little time, and demonstrates that this new family of tree regularizers yield models that are easier for humans to simulate than L1 or L2 penalties without sacrificing predictive power.
- Ensembles of Locally Independent Prediction Models33
This paper introduces a new diversity metric and associated method of training ensembles of models that extrapolate differently on local patches of the data manifold and finds that this method improves generalization and diversity in qualitatively novel ways, especially under data limits and covariate shift.
- Predicting change in diagnosis from major depression to bipolar disorder after antidepressant initiation30
Standard supervised machine learning methods enabled development of discriminative and transferable models to predict transition to bipolar disorder and these scores may enable physicians to more precisely calibrate follow-up intensity for high-risk patients after antidepressant initiation.
- Unsupervised Learning of PCFGs with Normalizing Flow29
A neural PCFG inducer which employs context embeddings (Peters et al., 2018) in a normalizing flow model to extend PCFG induction to use semantic and morphological information.
- 29
- Personalising AI Assistance Based on Overreliance Rate in AI-Assisted Decision Making28
This work considers a hidden quality previously identified as important: overreliance on AI assistance, and introduces a few probe questions to determine if a user is an overrelier or not, and improves human-AI team performance, personalising AI assistance based on users’ overreliance quality.
- Decomposition of Uncertainty for Active Learning and Reliable Reinforcement Learning in Stochastic Systems.27
This work shows how a decomposition of predictive uncertainty into its epistemic and aleatoric components arises naturally in a Bayesian active learning scenario and develops a new objective for reliable reinforcement learning (RL) with an epistemicandAleatoric risk element.
- Assessment of a Prediction Model for Antidepressant Treatment Stability Using Supervised Topic Models26
This cohort study assesses medication-specific treatment models vs general treatment models for predicting stable response to antidepressant treatment among patients with major depressive disorder.
- Agent Strategy Summarization26
A conceptual framework for strategy summarization is proposed, which is envisioning as a collaborative process that involves both agents and people, and possible testbeds that could be used to evaluate progress in research on strategy summaryization are suggested.
- 26
- 26
- 25
- An interpretable RL framework for pre-deployment modeling in ICU hypotension management23
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 and demonstrates the value of this approach via application to hypotension management in the ICU.
- Efficiently identifying individuals at high risk for treatment resistance in major depressive disorder using electronic health records22
Electronic health records facilitated stratification of risk for treatment-resistant depression and demonstrated generalizability to a second health system.
- Machine Learning Techniques for Accountability22
This short overview article is to begin the process of mapping the categories of methods that one could use to assess whether an AI system is meeting its objectives, and will not focus on any particular objective (such as safety, fairness, or robustness).
- Why do universal adversarial attacks work on large language models?: Geometry might be the answer21
By attacking the 117M parameter GPT-2 model, this work finds evidence indicating that universal adversarial triggers could be embedding vectors which merely approximate the semantic information in their adversarial training region.
- Assessing topic model relevance: Evaluation and informative priors21
This work proposes an additional topic quality metric that targets the stopword problem, and shows that it, unlike the standard measures, correctly correlates with human judgments of quality as defined by concentration of information‐rich words.
- 21
- 20
- Heterogeneity in Antidepressant Treatment and Major Depressive Disorder Outcomes Among Clinicians19
Groups of clinicians treating individuals diagnosed with major depressive disorder exhibit marked differences in prescribing patterns as well as longitudinal patient outcomes defined by electronic health records, suggesting the importance of considering treatment context in efforts at risk stratification.
- Improving Safety and Operational Efficiency in Residential Care Settings With WiFi-Based Localization18
Using timely resident location information can provide significant gains for both operational efficiency and enhanced resident safety (outdoor alerts); this approach may provide an inexpensive alternative for facilities that have sufficient wireless infrastructure.
- Interpretable Batch IRL to Extract Clinician Goals in ICU Hypotension Management.17
A novel interpretable batch variant of Adversarial Inverse Reinforcement Learning algorithm is developed that not only learns rewards which induce treatment policies similar to clinical treatments, but also ensures that the learned functional form of rewards is consistent with the decision mechanisms of clinicians in the ICU.
- 16
- 16
- Generalization in Clinical Prediction Models: The Blessing and Curse of Measurement Indicator Variables15
It is found that practice-specific features denoting measurement recency improved local performance and also led to more generalizable models if they are adjusted for during model development but discarded at validation.
- Depth-bounding is effective: Improvements and evaluation of unsupervised PCFG induction15
This work applies depth bounds within a chart-based Bayesian PCFG inducer, where bounding can be switched on and off, and then samples trees with or without bounding, showing that depth-bounding is indeed significantly effective in limiting the search space of the inducer and thereby increasing accuracy of resulting parsing model.
- 14
- The Indian Buffet Process: Scalable Inference and Extensions14
The core contribution of this thesis are three new inference procedures that allow inference in the IBP to be scaled from a few hundred to 100,000 observations, providing a general framework for correlated non-parametric feature models.
- Preference-based assistance optimization for lifting and lowering with a soft back exosuit13
Overall, participants preferred asymmetric parameter configurations with more lifting than lowering assistance, and that preferences were sensitive to user anthropometrics, highlighting the potential of perceptually guided assistance optimization for wearable robotic devices.
- Estimating Upper Extremity Fugl-Meyer Assessment Scores From Reaching Motions Using Wearable Sensors13
- Hidden Parameter Markov Decision Processes: An Emerging Paradigm for Modeling Families of Related Tasks.13
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).
- 12
- 12
- Big Data in the Assessment of Pediatric Medication Safety12
Big data (BD) in pediatric medication safety research provides many opportunities to improve the safety and health of children, and large distributed databases that include pediatric-specific outcome variables are available.
- Machine learning approaches to environmental disturbance rejection in multi-axis optoelectronic force sensors12
This work fabricates 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 and assesses viability of each algorithm tested in terms of both prediction accuracy and computational overhead.
- Cost-Sensitive Batch Mode Active Learning: Designing Astronomical Observation by Optimizing Telescope Time and Telescope Choice11
This work presents a batch, cost-sensitive, active learning approach that exploits structure in the unlabeled dataset, accounts for label uncertainty, and minimizes annotation costs and helps astronomers design effective experiments.
- 10
- Prediction Focused Topic Models via Feature Selection10
This work introduces a novel approach, the prediction-focused topic model, that uses the supervisory signal to retain only vocabulary terms that improve, or at least do not hinder, prediction performance.
- 10
- 10
- Latent Projection BNNs: Avoiding weight-space pathologies by learning latent representations of neural network weights.10
A novel variational inference framework for Bayesian neural networks that encodes complex distributions in high-dimensional parameter space with representations in a low-dimensional latent space, and performs inference efficiently on the low- dimensional representations is introduced.
- Improving counterfactual reasoning with kernelised dynamic mixing models10
This work presents a novel simulation-based reinforcement learning approach that mixes between models and kernel-based approaches to make its forward predictions, and demonstrates that it both learns state-of-the-art treatment policies and can make accurate forward predictions about the effects of treatments on unseen patients.
- Transfer Learning from Well-Curated to Less-Resourced Populations with HIV9
This work presents a novel mixture based approach that uses a deep information bottleneck to transfer patterns learned from European HIV cohorts—where genomic data is readily available—to African patients where no such data is available.
- Genomic Drivers of Coronary Artery Disease and Risk of Future Outcomes After Coronary Angiography8
It is suggested that genomic drivers of CAD are associated with the atherosclerosis burden observed angiographically and portend the risk of adverse outcomes.
- Depth-Bounded Statistical PCFG Induction as a Model of Human Grammar Acquisition8
An implementation of these memory bounds as limits on center embedding in a depth-specific transform of a recursive grammar yields a significant improvement over an equivalent but unbounded baseline, suggesting that this arrangement may indeed confer a learning advantage.
- 8
- 8
- 8
- 8
- Cross-corpora unsupervised learning of trajectories in autism spectrum disorders8
A fully unsupervised approach for learning disease trajectories from incomplete medical records and social media posts is presented, which uses a dynamic topic model approach which embeds each disease trajectory as a path in RD.
- A Comparison of Human and Agent Reinforcement Learning in Partially Observable Domains8
This work compares the performance of human subjects to RL techniques and finds that context—the meaning- fulness of the observations—plays a significant role in the rate of human RL, and shows that RL methods often learn faster and achieve better solu- tions than humans.
- Defining admissible rewards for high-confidence policy evaluation in batch reinforcement learning7
The approach to reward design on synthetic domains as well as in a critical care context is demonstrated, to guide the design of a reward function that consolidates clinical objectives to learn a policy for weaning patients from mechanical ventilation.
- 7
- 7
- 7
- 7
- Reinforcement Learning Interventions on Boundedly Rational Human Agents in Frictionful Tasks6
This paper introduces Behavior Model Reinforcement Learning (BMRL), a framework in which an AI agent intervenes on the parameters of a Markov Decision Process (MDP) belonging to a boundedly rational human agent, and introduces a notion of MDP equivalence specific to BMRL.
- The Unintended Consequences of Discount Regularization: Improving Regularization in Certainty Equivalence Reinforcement Learning6
An alternate view of discount regularization is revealed that exposes unintended consequences and it is demonstrated that planning under a lower discount factor produces an identical optimal policy to planning using any prior on the transition matrix that has the same distribution for all states and actions.
- Do clinicians follow heuristics in prescribing antidepressants?6
The 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.
- 6
- 6
- Challenges in Computing and Optimizing Upper Bounds of Marginal Likelihood based on Chi-Square Divergences6
It is shown that CHIVI is far more sensitive to initialization than classic VI based on KL minimization, often needs a very large number of samples, and may not be a reliable upper bound, and possible ways to detect and alleviate some of these pathologies are suggested.
- 6
- Connecting Federated ADMM to Bayes5
The work shows connection between two fields that are believed to be fundamentally different and combines them to improve federated learning and derive two versions of ADMM from VB that use flexible covariances and functional regularisation, respectively.
- AI Procurement Checklists: Revisiting Implementation in the Age of AI Governance5
Three key pitfalls around expertise, risk frameworks and transparency, that can decrease the efficacy of regulations aimed at government AI use are detailed and suggest avenues for improvement are suggested.
- XAI-Lyricist: Improving the Singability of AI-Generated Lyrics with Prosody Explanations5
XAI-Lyricist, leveraging musical prosody to guide LMs in generating singable lyrics and providing human-understandable singability explanations, shows that musical prosody can significantly improve the singability of LM-generated lyrics.
- Monitoring Fidelity of Online Reinforcement Learning Algorithms in Clinical Trials5
This paper proposes algorithm fidelity as a critical requirement for deploying online RL algorithms in clinical trials and presents a framework for pre-deployment planning and real-time monitoring to help algorithm developers and clinical researchers ensure algorithm fidelity.
- 5
- 5
- 5
- 5
- 5
- 5
- Restricted Indian buffet processes5
This work demonstrates how the restricted Indian buffet process circumvents this restriction, allowing arbitrary distributions over the number of features in an observation, and discusses several alternative constructions of the model.
- A Bayesian nonparametric approach to modeling battery health5
This work takes a data-driven approach and aims to learn a model of battery time-to-death from training data and uses a Dirichlet process prior over mixture weights to learn an infinite mixture model for battery health.
- Federated ADMM from Bayesian Duality4
It is shown that the solutions of variational-Bayesian (VB) objectives are associated with a duality structure that not only resembles the structure of ADMM's fixed-points but also generalizes it.
- Towards Regulatable AI Systems: Technical Gaps and Policy Opportunities4
This work investigates to what extent can AI experts vet an AI system for adherence to regulatory requirements through the lens of two public sector procurement checklists, identifying what can be done now, what should be possible with technical innovation, and what requirements need a more interdisciplinary approach.
- Modeling Mobile Health Users as Reinforcement Learning Agents4
This work formalizes the relationship between the user and mHealth with a framework in which the user optimizes a (potentially impaired) Markov Decision Process (MDP) and the mHealth agent intervenes on the user's MDP parameters.
- What Makes a Good Explanation?: A Harmonized View of Properties of Explanations4
Survey properties defined in interpretable machine learning papers, synthesize them based on what they actually measure, and describe the trade-offs between different formulations of these properties to enable more informed selection of task-appropriate formulations of explanation properties as well as standardization for future work in interpretability machine learning.
- Is Deep Reinforcement Learning Ready for Practical Applications in Healthcare? A Sensitivity Analysis of Duel-DDQN for Sepsis Treatment4
A sensitivity analysis on a state-of-the-art RL algorithm applied to hemodynamic stabilization treatment strategies for septic patients in the ICU finds that varying these settings can significantly impact learned policies, which suggests a need for caution when interpreting RL agent output.
- 4
- Prediction Focused Topic Models via Vocab Selection.4
A novel approach is introduced, the prediction-focused topic model, that uses the supervisory signal to retain only vocabulary terms that improve, or do not hinder, prediction performance, and is able to learn task-relevant, interpretable topics.
- Promoting Domain-Specific Terms in Topic Models with Informative Priors.4
This work proposes a simple strategy for automatically promoting terms with domain relevance and demoting these domain-specific stop words, and increases the amount of domain-relevant content and reduces the appearance of canonical and humanevaluated stopwords in three very different domains.
- 4
- 4
- Bayesian nonparametric approaches for reinforcement learning in partially observable domains4
It is shown how the representations learned using Bayesian nonparametric methods result in better performance and interesting learned structure in three contexts related to reinforcement learning in partially-observable domains: learning partially observable Markov Decision processes, taking advantage of expert demonstrations, and learning complex hidden structures such as dynamic Bayesian networks.
- 3
- Decision-Point Guided Safe Policy Improvement3
Decision Points RL (DPRL), an algorithm that restricts the set of state-action pairs considered for improvement, ensures high-confidence improvement in densely visited states (i.e. decision points) while still utilizing data from sparsely visited states.
- Identifying data-driven subtypes of major depressive disorder with electronic health records3
These data-driven subtypes of major depressive disorder, drawing on representative clinical cohorts, merit further investigation for their utility in identifying more homogeneous patient populations for basic as well as clinical investigation.
- 3
- Toward Computationally Efficient Inverse Reinforcement Learning via Reward Shaping3
This work motivates the use of potential-based reward shaping to reduce the computational burden of each RL sub-problem and serves as a proof-of-concept that will inspire future developments towards computationally efficient IRL.
- Discovering User Types: Mapping User Traits by Task-Specific Behaviors in Reinforcement Learning3
It is shown that seemingly different real-world environments admit the same set of user types and formalize this observation as an equivalence relation defined on environments.
- An Empirical Analysis of the Advantages of Finite- v.s. Infinite-Width Bayesian Neural Networks3
It is found that when the model is mis-specified, increasing width can hurt BNN performance, and there is evidence that finite-width BNNs generalize better partially due to the properties of their frequency spectrum that allows them to adapt under model mismatch.
- Policy Optimization with Sparse Global Contrastive Explanations3
This work develops a Reinforcement Learning (RL) framework for improving an existing behavior policy via sparse, user-interpretable changes with the constraint of keeping that global contrastive explanation short.
- 3
- 3
- Learned Uncertainty-Aware (LUNA) Bases for Bayesian Regression using Multi-Headed Auxiliary Networks.3
It is shown that traditional training procedures for NLMs can drastically underestimate uncertainty in data-scarce regions and a novel training procedure is proposed for capturing useful predictive uncertainties.
- 3
- Transfer Learning by Discovering Latent Task Parametrizations3
A framework that is able to discover the latent factors that parametrize a family of related tasks from data is presented, allowing an agent to flexibly adapt to task variations.
- A Benchmark for Multi-Party Negotiation Games from Real Negotiation Data2
A benchmark and evaluation framework for this setting is introduced, combining a configurable negotiation game generator with document-grounded instances derived from a climate negotiation exercise, showing that no solver dominates across regimes; performance depends on the structural properties of the game.
- Towards Integrating Personal Knowledge into Test-Time Predictions2
This work introduces the problem of human feature integration, which provides a way to incorporate important personal-knowledge from users without domain expertise into ML predictions.
- A Sim2Real Approach for Identifying Task-Relevant Properties in Interpretable Machine Learning2
XAIsim2real, a pipeline for running synthetic user studies that accurately predicts user preferences across three tasks and uncovers nuanced relationships, like how cognitive budget limits a user's engagement with complex explanations -- a trend confirmed with real users.
- Effective Monitoring of Online Decision-Making Algorithms in Digital Intervention Implementation2
Monitoring guidelines and findings give digital intervention teams the confidence to include online decision-making algorithms in digital interventions with the goal of safeguarding individuals and ensuring data quality.
- Travel-time prediction using neural-network-based mixture models2
A neural network-based mixture distributions to predict a user’s travel times given their origin and destination coordinates is proposed, showing that the method correctly estimates the travel time distribution, maximizes utility in a downstream elevator scheduling task, and is easy to retrain—making it a versatile and an inexpensive-to-maintain module when deployed in smart crowd management systems.
- 2
- 2
- Towards Robust Off-Policy Evaluation via Human Inputs2
This work proposes a novel framework, Robust OPE (ROPE), which considers shifts on a subset of covariates in the data based on user inputs, and estimates worst-case utility under these shifts, and develops computationally efficient algorithms that are robust to the aforementioned shifts for contextual bandits and Markov decision processes.
- 2
- Learning Deep Bayesian Latent Variable Regression Models that Generalize: When Non-identifiability is a Problem.2
This work develops 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, and improves upon benchmark methods across a range of synthetic and real datasets.
- 2
- Brief mentalizing imagery therapy reduces stress and enhances positive psychological traits in family dementia caregivers: mediation by mindfulness1
The impact of mentalizing imagery therapy (MIT), a mindfulness and guided imagery approach, on perceived stress and positive psychological traits and the role of dispositional mindfulness was investigated.
- Association between prescriber practices and major depression treatment outcomes1
Primary care and psychiatry predictors are not the same and therefore suggest potentially distinct strategies to improve clinical outcomes in each setting.
- Inverse Reinforcement Learning with Multiple Planning Horizons1
This work develops algorithms that can learn a global multi-agent reward function with agent-specific discount factors that reconstruct the expert policies and characterize the feasible solution space of the reward function and discount factors for both algorithms and demonstrate the generalizability of the learned reward function across multiple domains.
- On the Effective Horizon of Inverse Reinforcement Learning1
This analysis provides a guide for the principled choice of the effective horizon for IRL and prompts us to re-examine the classic IRL formulation: it is more natural to learn jointly the reward and the effective horizon rather than the reward alone with a given horizon.
- 1
- Signature Activation: A Sparse Signal View for Holistic Saliency1
Signature Activation is introduced, a saliency method that generates holistic and class-agnostic explanations for Convolutional Neural Network outputs and exploits the fact that certain kinds of medical images, such as angiograms, have clear foreground and background objects.
- (When) Are Contrastive Explanations of Reinforcement Learning Helpful?1
It is found 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 were larger, suggesting that contrastive explanations are not sufficient to solve the problem of effectively explaining reinforcement learning policies.
- Success of Uncertainty-Aware Deep Models Depends on Data Manifold Geometry1
It is found that the geometry of the data sub-manifold is an important factor in determining the success of various models, suggesting an interesting direction in the study of uncertainty-aware deep learning models.
- A Bayesian Approach to Learning Bandit Structure in Markov Decision Processes1
This work develops an online algorithm that uses a Bayesian hypothesis testing approach to learn the nature of the environment, and 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.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- PoRB-Nets: Poisson Process Radial Basis Function Networks1
A novel prior over radial basis function networks (RBFNs) is presented that allows for independent specification of functional amplitude variance and lengthscale (i.e., smoothness), where the inverse length-scale corresponds to the concentration of radial basis functions.
- 1
- 1
- 1
- 1
- 1
- Reports of the AAAI 2011 Spring Symposia1
This report summarizes the eight symposia of the 2011 Spring Symposium Series, presented by the Association for the Advancement of Artificial Intelligence at Stanford University.
- –
- –
- –
- EHR-based machine learning for psychiatric outcomes in youth and young adults: Challenges and opportunities–
A focused review of the small but growing body of work applying ML to EHR data across childhood and adolescence in the context of this adult literature highlights efforts to augment structured EHR data with dimensional and clinically-relevant constructs based on natural language processing (NLP) and patient-reported outcomes.
- –
- Prompting the Unknown: Understanding Response Uncertainty in Large Language Models–
A novel prompt-response concept model is proposed that explains the relationship between the amount of task-relevant information provided in the prompt and the LLM-generated response uncertainty by identifying four sources of response uncertainty: prompt underspecification, model quality, task variability, and semantic redundancy.
- Crowd Simulation Models for Enhanced Mass-Gathering Medical Response: A Practical Application–
This study marks a significant stride in demonstrating the potential of machine learning in crowd-egress prediction using video data by training a crowd simulation model on video data and proving validity by comparing its predictions to an actual crowd-egress event.
- Strategically Linked Decisions in Long-Term Planning and Reinforcement Learning–
This paper quantifies dependencies between planned actions with strategic link scores: the drop in the likelihood of one decision under the constraint that a follow-up decision is no longer available, and demonstrates the utility of strategic link scores through three practical applications.
- –
- –
- Guarantee Regions for Local Explanations–
This work proposes an anchor-based algorithm for identifying regions in which local explanations are guaranteed to be correct by explicitly describing those intervals along which the input features can be trusted.
- Towards Model-Agnostic Posterior Approximation for Fast and Accurate Variational Autoencoders–
It is shown that a deterministic, model-agnostic posterior approximation (MAPA) of the true model's posterior can be computed and used to develop a proof-of-concept inference method.
- –
- –
- –
- Diverse Concept Proposals for Concept Bottleneck Models–
This work identifies a number of predictive concepts that explain the data by offering multiple alternative explanations and allows the human expert to choose the one that best aligns with their expectation.
- –
- Bayesian Inverse Transition Learning for Offline Settings–
This work proposes a new constraint-based approach that captures the authors' desiderata for reliably learning a posterior distribution of the transition dynamics that is free from gradients and explains how combining uncertainty estimation with these constraints can help us infer a partial ranking of actions that produce higher returns.
- SAP-sLDA: An Interpretable Interface for Exploring Unstructured Text–
A semi-supervised human-in-the-loop LDA-based method for learning topics that preserve semantically meaningful relationships between documents in low-dimensional projections of text corpora yields more interpretable projections than baseline methods with only a fraction of labels provided.
- –
- –
- –
- –
- –
- Discussion–
This work presents a meta-analyses of Bayesian variable selection using non-local prior densities in ultrahigh-dimensional settings using a scalable MCMC algorithm for the horseshoe prior.
- –
- –
- –
- 1185-P: Predictive Model for Hyperglycemic Events after High Dose Corticosteroid Administration–
Several labs in the 24 hours prior to receipt of high dose steroids are associated with hyperglycemia, and the presence of a lactate result was the most significant predictor of future hyperglycemic events.
- –
- –
- –
- –
- –
- Nonparametric Bayesian Approaches for Reinforcement Learning in Partially Observable Domains–
The objective of the doctoral research is bring together two fields: partially-observable reinforcement learning (PORL) and non-parametric Bayesian statistics (NPB) to address issues of statistical modeling and decision-making in complex, real-world domains.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.