Hyewon Jeong

Hyewon Jeong, Collin M. Stultz, Marzyeh Ghassemi

Heart failure is a debilitating condition that affects millions of people worldwide and has a significant impact on their quality of life and mortality rates. An objective assessment of cardiac pressures remains an important method for the diagnosis and treatment prognostication for patients with heart failure. Although cardiac catheterization is the gold standard for estimating central hemodynamic pressures, it is an invasive procedure that carries inherent risks, making it a potentially dangerous procedure for some patients. Approaches that leverage non-invasive signals - such as electrocardiogram (ECG) - have the promise to make the routine estimation of cardiac pressures feasible in both inpatient and outpatient settings. Prior models trained to estimate intracardiac pressures (e.g., mean pulmonary capillary wedge pressure (mPCWP)) in a supervised fashion have shown good discriminatory ability but have been limited to the labeled dataset from the heart failure cohort. To address this issue and build a robust representation, we apply deep metric learning (DML) and propose a novel self-supervised DML with distance-based mining that improves the performance of a model with limited labels. We use a dataset that contains over 5.4 million ECGs without concomitant central pressure labels to pre-train a self-supervised DML model which showed improved classification of elevated mPCWP compared to self-supervised contrastive baselines. Additionally, the supervised DML model that uses ECGs with access to 8,172 mPCWP labels demonstrated significantly better performance on the mPCWP regression task compared to the supervised baseline. Moreover, our data suggest that DML yields models that are performant across patient subgroups, even when some patient subgroups are under-represented in the dataset. Our code is available at https://github.com/mandiehyewon/ssldml

Abinitha Gourabathina, Hyewon Jeong, Teya Bergamaschi et al. · mit

Deep learning models trained to optimize average accuracy often exhibit systematic failures on particular subpopulations. In real world settings, the subpopulations most affected by such disparities are frequently unlabeled or unknown, thereby motivating the development of methods that are performant on sensitive subgroups without being pre-specified. However, existing group-robust methods typically assume prior knowledge of relevant subgroups, using group annotations for training or model selection. We propose Low-rank Error Informed Adaptation (LEIA), a simple two-stage method that improves group robustness by identifying a low-dimensional subspace in the representation space where model errors concentrate. LEIA restricts adaptation to this error-informed subspace via a low-rank adjustment to the classifier logits, directly targeting latent failure modes without modifying the backbone or requiring group labels. Using five real-world datasets, we analyze group robustness under three settings: (1) truly no knowledge of subgroup relevance, (2) partial knowledge of subgroup relevance, and (3) full knowledge of subgroup relevance. Across all settings, LEIA consistently improves worst-group performance while remaining fast, parameter-efficient, and robust to hyperparameter choice.

Yubin Kim, Chanwoo Park, Hyewon Jeong et al. · mit

Foundation models are becoming valuable tools in medicine. Yet despite their promise, the best way to leverage Large Language Models (LLMs) in complex medical tasks remains an open question. We introduce a novel multi-agent framework, named Medical Decision-making Agents (MDAgents) that helps address this gap by automatically assigning a collaboration structure to a team of LLMs. The assigned solo or group collaboration structure is tailored to the medical task at hand, emulating real-world medical decision-making processes adapted to tasks of varying complexities. We evaluate our framework and baseline methods using state-of-the-art LLMs across a suite of real-world medical knowledge and medical diagnosis benchmarks, including a comparison of LLMs' medical complexity classification against human physicians. MDAgents achieved the best performance in seven out of ten benchmarks on tasks requiring an understanding of medical knowledge and multi-modal reasoning, showing a significant improvement of up to 4.2% (p < 0.05) compared to previous methods' best performances. Ablation studies reveal that MDAgents effectively determines medical complexity to optimize for efficiency and accuracy across diverse medical tasks. Notably, the combination of moderator review and external medical knowledge in group collaboration resulted in an average accuracy improvement of 11.8%. Our code can be found at https://github.com/mitmedialab/MDAgents.

Yeongwoong Kim, Hyewon Jeong, Janghyun Yu et al.

The proliferation of deep learning-based machine vision applications has given rise to a new type of compression, so called video coding for machine (VCM). VCM differs from traditional video coding in that it is optimized for machine vision performance instead of human visual quality. In the feature compression track of MPEG-VCM, multi-scale features extracted from images are subject to compression. Recent feature compression works have demonstrated that the versatile video coding (VVC) standard-based approach can achieve a BD-rate reduction of up to 96% against MPEG-VCM feature anchor. However, it is still sub-optimal as VVC was not designed for extracted features but for natural images. Moreover, the high encoding complexity of VVC makes it difficult to design a lightweight encoder without sacrificing performance. To address these challenges, we propose a novel multi-scale feature compression method that enables both the end-to-end optimization on the extracted features and the design of lightweight encoders. The proposed model combines a learnable compressor with a multi-scale feature fusion network so that the redundancy in the multi-scale features is effectively removed. Instead of simply cascading the fusion network and the compression network, we integrate the fusion and encoding processes in an interleaved way. Our model first encodes a larger-scale feature to obtain a latent representation and then fuses the latent with a smaller-scale feature. This process is successively performed until the smallest-scale feature is fused and then the encoded latent at the final stage is entropy-coded for transmission. The results show that our model outperforms previous approaches by at least 52% BD-rate reduction and has $\times5$ to $\times27$ times less encoding time for object detection...

Haoran Zhang, Aparna Balagopalan, Nassim Oufattole et al.

Large repositories of image-caption pairs are essential for the development of vision-language models. However, these datasets are often extracted from noisy data scraped from the web, and contain many mislabeled instances. In order to improve the reliability of downstream models, it is important to identify and filter images with incorrect captions. However, beyond filtering based on image-caption embedding similarity, no prior works have proposed other methods to filter noisy multimodal data, or concretely assessed the impact of noisy captioning data on downstream training. In this work, we propose, theoretically justify, and empirically validate LEMoN, a method to identify label errors in image-caption datasets. Our method leverages the multimodal neighborhood of image-caption pairs in the latent space of contrastively pretrained multimodal models to automatically identify label errors. Through empirical evaluations across eight datasets and twelve baselines, we find that LEMoN outperforms the baselines by over 3% in label error detection, and that training on datasets filtered using our method improves downstream captioning performance by more than 2 BLEU points over noisy training.

Xiao Xiang, David Restrepo, Hyewon Jeong et al.

Learning from electronic health records (EHRs) time series is challenging due to irregular sam- pling, heterogeneous missingness, and the resulting sparsity of observations. Prior self-supervised meth- ods either impute before learning, represent missingness through a dedicated input signal, or optimize solely for imputation, reducing their capacity to efficiently learn representations that support clinical downstream tasks. We propose the Augmented-Intrinsic Dual-Masked Autoencoder (AID-MAE), which learns directly from incomplete time series by applying an intrinsic missing mask to represent naturally missing values and an augmented mask that hides a subset of observed values for reconstruction during training. AID-MAE processes only the unmasked subset of tokens and consistently outperforms strong baselines, including XGBoost and DuETT, across multiple clinical tasks on two datasets. In addition, the learned embeddings naturally stratify patient cohorts in the representation space.

Hyewon Jeong, Suyeol Yun, Hammaad Adam

Electrocardiograms (ECGs) are an established technique to screen for abnormal cardiac signals. Recent work has established that it is possible to detect arrhythmia directly from the ECG signal using deep learning algorithms. While a few prior approaches with contrastive learning have been successful, the best way to define a positive sample remains an open question. In this project, we investigate several ways to define positive samples, and assess which approach yields the best performance in a downstream task of classifying arrhythmia. We explore spatiotemporal invariances, generic augmentations, demographic similarities, cardiac rhythms, and wave attributes of ECG as potential ways to match positive samples. We then evaluate each strategy with downstream task performance, and find that learned representations invariant to patient identity are powerful in arrhythmia detection. We made our code available in: https://github.com/mandiehyewon/goodviews_ecg.git

Kwanhyung Lee, Hyewon Jeong, Seyun Kim et al.

Electroencephalogram (EEG) is an important diagnostic test that physicians use to record brain activity and detect seizures by monitoring the signals. There have been several attempts to detect seizures and abnormalities in EEG signals with modern deep learning models to reduce the clinical burden. However, they cannot be fairly compared against each other as they were tested in distinct experimental settings. Also, some of them are not trained in real-time seizure detection tasks, making it hard for on-device applications. Therefore in this work, for the first time, we extensively compare multiple state-of-the-art models and signal feature extractors in a real-time seizure detection framework suitable for real-world application, using various evaluation metrics including a new one we propose to evaluate more practical aspects of seizure detection models. Our code is available at https://github.com/AITRICS/EEG_real_time_seizure_detection.

A. Tuan Nguyen, Hyewon Jeong, Eunho Yang et al.

Although recent multi-task learning methods have shown to be effective in improving the generalization of deep neural networks, they should be used with caution for safety-critical applications, such as clinical risk prediction. This is because even if they achieve improved task-average performance, they may still yield degraded performance on individual tasks, which may be critical (e.g., prediction of mortality risk). Existing asymmetric multi-task learning methods tackle this negative transfer problem by performing knowledge transfer from tasks with low loss to tasks with high loss. However, using loss as a measure of reliability is risky since it could be a result of overfitting. In the case of time-series prediction tasks, knowledge learned for one task (e.g., predicting the sepsis onset) at a specific timestep may be useful for learning another task (e.g., prediction of mortality) at a later timestep, but lack of loss at each timestep makes it difficult to measure the reliability at each timestep. To capture such dynamically changing asymmetric relationships between tasks in time-series data, we propose a novel temporal asymmetric multi-task learning model that performs knowledge transfer from certain tasks/timesteps to relevant uncertain tasks, based on feature-level uncertainty. We validate our model on multiple clinical risk prediction tasks against various deep learning models for time-series prediction, which our model significantly outperforms, without any sign of negative transfer. Further qualitative analysis of learned knowledge graphs by clinicians shows that they are helpful in analyzing the predictions of the model. Our final code is available at https://github.com/anhtuan5696/TPAMTL.

Yubin Kim, Hyewon Jeong, Shan Chen et al.

Hallucinations in foundation models arise from autoregressive training objectives that prioritize token-likelihood optimization over epistemic accuracy, fostering overconfidence and poorly calibrated uncertainty. We define medical hallucination as any model-generated output that is factually incorrect, logically inconsistent, or unsupported by authoritative clinical evidence in ways that could alter clinical decisions. We evaluated 11 foundation models (7 general-purpose, 4 medical-specialized) across seven medical hallucination tasks spanning medical reasoning and biomedical information retrieval. General-purpose models achieved significantly higher proportions of hallucination-free responses than medical-specialized models (median: 76.6% vs 51.3%, difference = 25.2%, 95% CI: 18.7-31.3%, Mann-Whitney U = 27.0, p = 0.012, rank-biserial r = -0.64). Top-performing models such as Gemini-2.5 Pro exceeded 97% accuracy when augmented with chain-of-thought prompting (base: 87.6%), while medical-specialized models like MedGemma ranged from 28.6-61.9% despite explicit training on medical corpora. Chain-of-thought reasoning significantly reduced hallucinations in 86.4% of tested comparisons after FDR correction (q < 0.05), demonstrating that explicit reasoning traces enable self-verification and error detection. Physician audits confirmed that 64-72% of residual hallucinations stemmed from causal or temporal reasoning failures rather than knowledge gaps. A global survey of clinicians (n = 70) validated real-world impact: 91.8% had encountered medical hallucinations, and 84.7% considered them capable of causing patient harm. The underperformance of medical-specialized models despite domain training indicates that safety emerges from sophisticated reasoning capabilities and broad knowledge integration developed during large-scale pre-training, not from narrow optimization.

Aritra Roy, Kevin Shen, Andrew MacBride et al.

Large language models (LLMs) are rapidly changing how researchers in materials science and chemistry discover, organize, and act on scientific knowledge. This paper analyzes a broad set of community-developed LLM applications in an effort to identify emerging patterns in how these systems can be used across the scientific research lifecycle. We organize the projects into two complementary categories: Knowledge Infrastructure, systems that structure, retrieve, synthesize, and validate scientific information; and Action Systems, systems that execute, coordinate, or automate scientific work across computational and experimental environments. The submissions reveal a shift from single-purpose LLM tools toward integrated, multi-agent workflows that combine retrieval, reasoning, tool use, and domain-specific validation. Prominent themes include retrieval-augmented generation as grounding infrastructure, persistent structured knowledge representations, multimodal and multilingual scientific inputs, and early progress toward laboratory-integrated closed-loop systems. Together, these results suggest that LLMs are evolving from general-purpose assistants into composable infrastructure for scientific reasoning and action. This work provides a community snapshot of that transition and a practical taxonomy for understanding emerging LLM-enabled workflows in materials science and chemistry.

Maxwell A. Xu, Jaya Narain, Gregory Darnell et al.

We present RelCon, a novel self-supervised Relative Contrastive learning approach for training a motion foundation model from wearable accelerometry sensors. First, a learnable distance measure is trained to capture motif similarity and domain-specific semantic information such as rotation invariance. Then, the learned distance provides a measurement of semantic similarity between a pair of accelerometry time-series, which we use to train our foundation model to model relative relationships across time and across subjects. The foundation model is trained on 1 billion segments from 87,376 participants, and achieves state-of-the-art performance across multiple downstream tasks, including human activity recognition and gait metric regression. To our knowledge, we are the first to show the generalizability of a foundation model with motion data from wearables across distinct evaluation tasks.

Hyewon Jeong, Nassim Oufattole, Matthew Mcdermott et al.

In clinical practice, one often needs to identify whether a patient is at high risk of adverse outcomes after some key medical event. For example, quantifying the risk of adverse outcomes after an acute cardiovascular event helps healthcare providers identify those patients at the highest risk of poor outcomes; i.e., patients who benefit from invasive therapies that can lower their risk. Assessing the risk of adverse outcomes, however, is challenging due to the complexity, variability, and heterogeneity of longitudinal medical data, especially for individuals suffering from chronic diseases like heart failure. In this paper, we introduce Event-Based Contrastive Learning (EBCL) - a method for learning embeddings of heterogeneous patient data that preserves temporal information before and after key index events. We demonstrate that EBCL can be used to construct models that yield improved performance on important downstream tasks relative to other pretraining methods. We develop and test the method using a cohort of heart failure patients obtained from a large hospital network and the publicly available MIMIC-IV dataset consisting of patients in an intensive care unit at a large tertiary care center. On both cohorts, EBCL pretraining yields models that are performant with respect to a number of downstream tasks, including mortality, hospital readmission, and length of stay. In addition, unsupervised EBCL embeddings effectively cluster heart failure patients into subgroups with distinct outcomes, thereby providing information that helps identify new heart failure phenotypes. The contrastive framework around the index event can be adapted to a wide array of time-series datasets and provides information that can be used to guide personalized care.

Yuexing Hao, Kumail Alhamoud, Hyewon Jeong et al.

Large Language Models (LLMs) have demonstrated remarkable performance on various medical question-answering (QA) benchmarks, including standardized medical exams. However, correct answers alone do not ensure correct logic, and models may reach accurate conclusions through flawed processes. In this study, we introduce the MedPAIR (Medical Dataset Comparing Physicians and AI Relevance Estimation and Question Answering) dataset to evaluate how physician trainees and LLMs prioritize relevant information when answering QA questions. We obtain annotations on 1,300 QA pairs from 36 physician trainees, labeling each sentence within the question components for relevance. We compare these relevance estimates to those for LLMs, and further evaluate the impact of these "relevant" subsets on downstream task performance for both physician trainees and LLMs. We find that LLMs are frequently not aligned with the content relevance estimates of physician trainees. After filtering out physician trainee-labeled irrelevant sentences, accuracy improves for both the trainees and the LLMs. All LLM and physician trainee-labeled data are available at: http://medpair.csail.mit.edu/.

Yubin Kim, Chanwoo Park, Hyewon Jeong et al.

Medical Decision-Making (MDM) is a multi-faceted process that requires clinicians to assess complex multi-modal patient data patient, often collaboratively. Large Language Models (LLMs) promise to streamline this process by synthesizing vast medical knowledge and multi-modal health data. However, single-agent are often ill-suited for nuanced medical contexts requiring adaptable, collaborative problem-solving. Our MDAgents addresses this need by dynamically assigning collaboration structures to LLMs based on task complexity, mimicking real-world clinical collaboration and decision-making. This framework improves diagnostic accuracy and supports adaptive responses in complex, real-world medical scenarios, making it a valuable tool for clinicians in various healthcare settings, and at the same time, being more efficient in terms of computing cost than static multi-agent decision making methods.

Hyewon Jeong, Sarah Jabbour, Yuzhe Yang et al. · uw

The third ML4H symposium was held in person on December 10, 2023, in New Orleans, Louisiana, USA. The symposium included research roundtable sessions to foster discussions between participants and senior researchers on timely and relevant topics for the \ac{ML4H} community. Encouraged by the successful virtual roundtables in the previous year, we organized eleven in-person roundtables and four virtual roundtables at ML4H 2022. The organization of the research roundtables at the conference involved 17 Senior Chairs and 19 Junior Chairs across 11 tables. Each roundtable session included invited senior chairs (with substantial experience in the field), junior chairs (responsible for facilitating the discussion), and attendees from diverse backgrounds with interest in the session's topic. Herein we detail the organization process and compile takeaways from these roundtable discussions, including recent advances, applications, and open challenges for each topic. We conclude with a summary and lessons learned across all roundtables. This document serves as a comprehensive review paper, summarizing the recent advancements in machine learning for healthcare as contributed by foremost researchers in the field.

Nassim Oufattole, Teya Bergamaschi, Aleksia Kolo et al. · mit

Effective, reliable, and scalable development of machine learning (ML) solutions for structured electronic health record (EHR) data requires the ability to reliably generate high-quality baseline models for diverse supervised learning tasks in an efficient and performant manner. Historically, producing such baseline models has been a largely manual effort--individual researchers would need to decide on the particular featurization and tabularization processes to apply to their individual raw, longitudinal data; and then train a supervised model over those data to produce a baseline result to compare novel methods against, all for just one task and one dataset. In this work, powered by complementary advances in core data standardization through the MEDS framework, we dramatically simplify and accelerate this process of tabularizing irregularly sampled time-series data, providing researchers the ability to automatically and scalably featurize and tabularize their longitudinal EHR data across tens of thousands of individual features, hundreds of millions of clinical events, and diverse windowing horizons and aggregation strategies, all before ultimately leveraging these tabular data to automatically produce high-caliber XGBoost baselines in a highly computationally efficient manner. This system scales to dramatically larger datasets than tabularization tools currently available to the community and enables researchers with any MEDS format dataset to immediately begin producing reliable and performant baseline prediction results on various tasks, with minimal human effort required. This system will greatly enhance the reliability, reproducibility, and ease of development of powerful ML solutions for health problems across diverse datasets and clinical settings.

Yubin Kim, Zhiyuan Hu, Hyewon Jeong et al.

Large Language Models (LLMs) as clinical agents require careful behavioral adaptation. While adept at reactive tasks (e.g., diagnosis reasoning), LLMs often struggle with proactive engagement, like unprompted identification of critical missing information or risks. We introduce BehaviorBench, a comprehensive dataset to evaluate agent behaviors across a clinical assistance spectrum, ranging from reactive query responses to proactive interventions (e.g., clarifying ambiguities, flagging overlooked critical data). Our BehaviorBench experiments reveal LLMs' inconsistent proactivity. To address this, we propose BehaviorSFT, a novel training strategy using behavioral tokens to explicitly condition LLMs for dynamic behavioral selection along this spectrum. BehaviorSFT boosts performance, achieving up to 97.3% overall Macro F1 on BehaviorBench and improving proactive task scores (e.g., from 95.0% to 96.5% for Qwen2.5-7B-Ins). Crucially, blind clinician evaluations confirmed BehaviorSFT-trained agents exhibit more realistic clinical behavior, striking a superior balance between helpful proactivity (e.g., timely, relevant suggestions) and necessary restraint (e.g., avoiding over-intervention) versus standard fine-tuning or explicit instructed agents.

Mahtab Bigverdi, Wisdom Ikezogwo, Kevin Zhang et al.

Multimodal language models (MLMs) show promise for clinical decision support and diagnostic reasoning, raising the prospect of end-to-end automated medical image interpretation. However, clinicians are highly selective in adopting AI tools; a model that makes errors on seemingly simple perception tasks such as determining image orientation or identifying whether a CT scan is contrast-enhance are unlikely to be adopted for clinical tasks. We introduce Medblink, a benchmark designed to probe these models for such perceptual abilities. Medblink spans eight clinically meaningful tasks across multiple imaging modalities and anatomical regions, totaling 1,429 multiple-choice questions over 1,605 images. We evaluate 19 state-of-the-art MLMs, including general purpose (GPT4o, Claude 3.5 Sonnet) and domain specific (Med Flamingo, LLaVA Med, RadFM) models. While human annotators achieve 96.4% accuracy, the best-performing model reaches only 65%. These results show that current MLMs frequently fail at routine perceptual checks, suggesting the need to strengthen their visual grounding to support clinical adoption. Data is available on our project page.

Yubin Kim, Hyewon Jeong, Chanwoo Park et al.

Large language models (LLMs) deployed as agents introduce significant safety risks in clinical settings due to their potential for error and single points of failure. We introduce Tiered Agentic Oversight (TAO), a hierarchical multi-agent system that enhances AI safety through layered, automated supervision. Inspired by clinical hierarchies (e.g., nurse-physician-specialist) in hospital, TAO routes tasks to specialized agents based on complexity, creating a robust safety framework through automated inter- and intra-tier communication and role-playing. Crucially, this hierarchical structure functions as an effective error-correction mechanism, absorbing up to 24% of individual agent errors before they can compound. Our experiments reveal TAO outperforms single-agent and other multi-agent systems on 4 out of 5 healthcare safety benchmarks, with up to an 8.2% improvement. Ablation studies confirm key design principles of the system: (i) its adaptive architecture is over 3% safer than static, single-tier configurations, and (ii) its lower tiers are indispensable, as their removal causes the most significant degradation in overall safety. Finally, we validated the system's synergy with human doctors in a user study where a physician, acting as the highest tier agent, provided corrective feedback that improved medical triage accuracy from 40% to 60%. Project Page: https://tiered-agentic-oversight.github.io/

Hyewon Jeong, Siddharth Nayak, Taylor Killian et al.

Sepsis is a life-threatening condition defined by end-organ dysfunction due to a dysregulated host response to infection. Although the Surviving Sepsis Campaign has launched and has been releasing sepsis treatment guidelines to unify and normalize the care for sepsis patients, it has been reported in numerous studies that disparities in care exist across the trajectory of patient stay in the emergency department and intensive care unit. Here, we apply a number of reinforcement learning techniques including behavioral cloning, imitation learning, and inverse reinforcement learning, to learn the optimal policy in the management of septic patient subgroups using expert demonstrations. Then we estimate the counterfactual optimal policies by applying the model to another subset of unseen medical populations and identify the difference in cure by comparing it to the real policy. Our data comes from the sepsis cohort of MIMIC-IV and the clinical data warehouses of the Mass General Brigham healthcare system. The ultimate objective of this work is to use the optimal learned policy function to estimate the counterfactual treatment policy and identify deviations across sub-populations of interest. We hope this approach would help us identify any disparities in care and also changes in cure in response to the publication of national sepsis treatment guidelines.

Jay Heo, Junhyeon Park, Hyewon Jeong et al.

We propose a novel interactive learning framework which we refer to as Interactive Attention Learning (IAL), in which the human supervisors interactively manipulate the allocated attentions, to correct the model's behavior by updating the attention-generating network. However, such a model is prone to overfitting due to scarcity of human annotations, and requires costly retraining. Moreover, it is almost infeasible for the human annotators to examine attentions on tons of instances and features. We tackle these challenges by proposing a sample-efficient attention mechanism and a cost-effective reranking algorithm for instances and features. First, we propose Neural Attention Process (NAP), which is an attention generator that can update its behavior by incorporating new attention-level supervisions without any retraining. Secondly, we propose an algorithm which prioritizes the instances and the features by their negative impacts, such that the model can yield large improvements with minimal human feedback. We validate IAL on various time-series datasets from multiple domains (healthcare, real-estate, and computer vision) on which it significantly outperforms baselines with conventional attention mechanisms, or without cost-effective reranking, with substantially less retraining and human-model interaction cost.