๐ฉบ Medical LLM¶
๐ง NeurIPS2026 ยท 16 paper notes
๐ Same area in other venues: ๐ท CVPR2026 (1) ยท ๐ฌ ICLR2026 (20) ยท ๐ฌ ACL2026 (47) ยท ๐งช ICML2026 (4) ยท ๐ค AAAI2026 (12) ยท ๐ง NeurIPS2025 (17)
๐ฅ Top topics: Medical Imaging ร8 ยท LLM ร2 ยท Reasoning ร2
- Donโt Pay Attention, PLANT It: Pretraining Attention via Learning-to-Rank
-
Addressing the failure of random label attention initialization on skewed long-tail classes in Extreme Multi-Label Text Classification, this paper introduces PLANT, a plug-and-play framework that pretrains label attention via Mutual Information Gain and Learning-to-Rank, substantially outperforming larger LLM baselines and prior SOTA models on rare labels.
- Ask, Answer, and Detect: Role-Playing LLMs for Personality Detection with Question-Conditioned Mixture-of-Experts
-
ROME establishes a questionnaire-grounded "Ask, Answer, and Detect" paradigm that leverages role-playing LLMs to simulate offline psychological questionnaire responses and uses a question-conditioned Mixture-of-Experts module to achieve interpretable, highly accurate social media personality detection.
- Deceptive Grounding: Entity Attribution Failure in Clinical Retrieval-Augmented Generation
-
Reveals the deceptive grounding phenomenon in clinical RAG where genuine retrieved clinical trial evidence is misattributed to queried drugs while evading standard faithfulness evaluations, reveals a two-stage permission gate mechanism, and demonstrates entity-attribution verification with 98.7% recall.
- Discrete Neural Interlingua for Ontology-Agnostic EHR Modeling
-
This paper introduces Mediator, a discrete ontology-agnostic EHR tokenizer that maps arbitrary clinical concept text into a shared vector-quantized (VQ) codebook to serve as a neural interlingua, overcoming closed-vocabulary bottlenecks and enabling zero-shot cross-hospital transfer.
- DoAtlas-1: A Causal Compilation Paradigm for Clinical AI
-
Introduces the causal compilation paradigm and the DoAtlas-1 system, which transforms heterogeneous medical evidence from narrative text into executable interventional estimand objects, supporting six typed causal queries with statistical closed-loop validation on the 10,000-participant Human Phenotype Project cohort.
- DT-PBO: an Interpretable Tree-based Surrogate Model for Preferential Bayesian Optimization
-
Proposes DT-PBO, the first non-linear inherently interpretable tree-based surrogate model for preferential Bayesian optimization, using an innovative pairwise consistency score and straddler exclusion to construct shallow trees with Laplace-approximated zero-sum probabilistic leaves, delivering competitive convergence, up to 400x faster runtime, and transparent clinical triage logic.
- EHRNote-ChatQA: A Benchmark for Evidence-Grounded Multi-Turn Clinical Question Answering over Longitudinal Discharge Summaries
-
EHRNote-ChatQA introduces the first evidence-grounded, multi-turn clinical question answering benchmark spanning patients' longitudinal discharge summaries (967 patients, 16,072 expert-verified QA pairs), showing that LLMs suffer a severe 22.0 pp drop in evidence grounding compared to content answering and exhibit compounding errors across dialogue turns.
- Extending ROC Analysis to Uncertainty-Aware Risk Prediction with an Interval-Based AUC (iAUC)
-
To address the inability of scalar scores to capture epistemic uncertainty in high-stakes clinical risk prediction, this paper extends ROC analysis to interval-valued predictions via the iAUC framework, constructing lower and upper ROC curves that induce a three-region ranking decomposition (confidently correct, ambiguous, and confidently incorrect) and establishing theoretical coverage-based bounds on the Bayes-optimal AUC.
- MedEvoEval: Evaluating Continual Evolution of Doctor Agents through Simulated Clinical Episodes
-
MedEvoEval introduces an executable longitudinal evaluation framework for doctor agents based on action-gated simulated outpatient episodes, revealing process costs hidden by final-answer scoring and demonstrating that accumulated rule memory improves interaction efficiency and composite utility on unseen held-out cases while preserving backward retention.
- MedKIT: Evaluating Knowledge Integration and Generalization in Large Language Models
-
MedKIT converts 6,196 timestamped oncology evidence updates into fact-centered, multi-task probes and compares 12 knowledge integration methods across 5 models, showing that recalling an updated fact usually does not imply using it in open-ended reasoning or generation.
- Medmarks: A Comprehensive Open-Source LLM Benchmark Suite for Medical Tasks
-
Medmarks introduces a fully open-source automated evaluation suite spanning 30 medical task benchmarks across verifiable tasks (Medmarks-V) and open-ended clinical reasoning (Medmarks-OE), systematically evaluating 61 language models across 71 configurations and packaging 8 train/test split environments as Medmarks-T to directly support post-training reinforcement learning.
- MedMemoryBench: Benchmarking Agent Memory in Personalized Healthcare
-
MedMemoryBench benchmarks agent memory in personalized healthcare by constructing year-long longitudinal patient event graphs with safety-critical trap events and introducing an evaluate-while-constructing streaming protocol, revealing that accumulated clinical noise triggers memory saturation and degrades multi-hop reasoning across mainstream memory architectures.
- MedMisBench: Measuring Epistemic Resilience of LLMs Under Misleading Medical Context
-
Addressing the critical flaw that standard medical benchmarks evaluate LLMs only under pristine conditions, this paper introduces MedMisBench to assess epistemic resilienceโpreserving correct medical judgment when adversarial context is introducedโacross 10,932 items and 48,889 misleading context-option pairs, revealing that focused rule-like falsehoods drop mean model accuracy from 71.1% to 38.0% with a 51.5% attack success rate and a 38.2% expert-reviewed serious clinical harm rate.
- MedZERO: Self-Evolving Agents for Open-Ended Medical Reasoning Through Controlled Knowledge Accumulation
-
MedZERO addresses open-ended, partially verifiable medical reasoning by co-evolving an Examiner and Reasoner through frontier curriculum generation and ephemeral-to-persistent knowledge accumulation, substantially breaking the self-evolution bottleneck without expert annotations.
- MoBayes: A Modular Bayesian Framework for Separating Reasoning from Language in Conversational Clinical Decision Support
-
MoBayes strictly confines large language models to parsing patient utterances into structured evidence and verbalizing questions, delegating posterior tracking, expected information gain-based question selection, and controllable abstention to a deterministic Bayesian engine that outperforms frontier monolithic clinical LLMs at a fraction of the cost.
- TreeWalker: Partial Evaluation for Grouped Tree-Ensemble Inference
-
TreeWalker presents an online partial evaluation engine for grouped tree-ensemble inference that treats shared features as static and varying features as dynamic, traversing each tree once per group with bitmask propagation to achieve 2.7รโ9.1ร algorithmic speedup and up to 54ร system speedup while preserving bit-exact numerical accuracy.