MAY WANG

J. Ben Tamo, Yuxing Lu, Benoit L. Marteau et al.

Large Language Models (LLMs) are fluent but prone to hallucinations, producing answers that appear plausible yet are unsupported by available evidence. This failure is especially problematic in high-stakes domains where decisions must be justified by verifiable information. We introduce \textbf{EvidenceRL}, a reinforcement learning framework that enforces evidence adherence during training. EvidenceRL scores candidate responses for grounding (entailment with retrieved evidence and context) and correctness (agreement with reference answers) and optimizes the generator using Group Relative Policy Optimization (GRPO). We evaluate across two high-stakes domains, cardiac diagnosis and legal reasoning, where EvidenceRL consistently improves evidence grounding and faithfulness without sacrificing task accuracy. On cardiac diagnosis, F1@3 increases from 37.0 to 54.5 on Llama-3.2-3B while grounding ($G_{\max}@3$) rises from 47.6 to 78.2; hallucinations drop nearly 5$\times$ and evidence-supported diagnoses increase from 31.8\% to 61.6\%. On legal reasoning, EvidenceRL raises Faithfulness from 32.8\% to 67.6\% on Llama-3.1-8B, demonstrating consistent behavioral change across domains. Our code is open-sourced at https://github.com/Wizaaard/EvidenceRL.git.

Micky C. Nnamdi, Benoit L. Marteau, Yishan Zhong et al.

Large Multimodal Models (LMMs) achieve state-of-the-art performance in high-stakes domains like healthcare, yet their reasoning remains opaque. Current interpretability methods, such as attention mechanisms or post-hoc saliency, often fail to faithfully represent the model's decision-making process, particularly when integrating heterogeneous modalities like time-series and text. We introduce Tree-of-Evidence (ToE), an inference-time search algorithm that frames interpretability as a discrete optimization problem. Rather than relying on soft attention weights, ToE employs lightweight Evidence Bottlenecks that score coarse groups or units of data (e.g., vital-sign windows, report sentences) and performs a beam search to identify the compact evidence set required to reproduce the model's prediction. We evaluate ToE across six tasks spanning three datasets and two domains: four clinical prediction tasks on MIMIC-IV, cross-center validation on eICU, and non-clinical fault detection on LEMMA-RCA. ToE produces auditable evidence traces while maintaining predictive performance, retaining over 0.98 of full-model AUROC with as few as five evidence units across all settings. Under sparse evidence budgets, ToE achieves higher decision agreement and lower probability fidelity error than other approaches. Qualitative analyses show that ToE adapts its search strategy: it often resolves straightforward cases using only vitals, while selectively incorporating text when physiological signals are ambiguous. ToE therefore provides a practical mechanism for auditing multimodal models by revealing which discrete evidence units support each prediction.

Aditya Shukla, Yining Yuan, Ben Tamo et al.

Large language models (LLMs) can generate fluent clinical summaries of remote therapeutic monitoring time series. However, it remains unclear whether these narratives faithfully capture clinically significant events, such as sustained abnormalities. Existing evaluation metrics primarily focus on semantic similarity and linguistic quality, leaving event-level correctness largely unmeasured. To address this gap, we introduce an event-based evaluation framework for multimodal time-series summarization using the Technology-Integrated Health Management (TIHM)-1.5 dementia monitoring dataset. Clinically grounded daily events are derived through rule-based abnormal thresholds and temporal persistence criteria. Model-generated summaries are then aligned with these structured facts. Our evaluation protocol measures abnormality recall, duration recall, measurement coverage, and hallucinated event mentions. We benchmark three approaches: zero-shot prompting, statistical prompting, and a vision-based pipeline that uses rendered time-series visualizations. The results reveal a striking decoupling between conventional metrics and clinical event fidelity. Models that achieve high semantic similarity scores often exhibit near-zero abnormality recall. In contrast, the vision-based approach demonstrates the strongest event alignment, achieving 45.7% abnormality recall and 100% duration recall. These findings underscore the importance of event-aware evaluation to ensure reliable clinical time-series summarization.

Jieru Li, Matthew Chen, Micky C. Nnamdi et al.

Medical image segmentation models built on Segment Anything Model (SAM) achieve strong performance on clean benchmarks, yet their reliability often degrades under realistic image corruptions such as noise, blur, motion artifacts, and modality-specific distortions. Existing approaches address either medical-domain adaptation or corruption robustness, but not both jointly. In SAM, we find that these capabilities are concentrated in complementary modules: the image encoder preserves medical priors, while the mask decoder governs corruption robustness. Motivated by this observation, we propose RobustMedSAM, which adopts module-wise checkpoint fusion by initializing the image encoder from MedSAM and the mask decoder from RobustSAM under a shared ViT-B architecture. We then fine-tune only the mask decoder on 35 medical datasets from MedSegBench, spanning six imaging modalities and 12 corruption types, while freezing the remaining components to preserve pretrained medical representations. We additionally investigate an SVD-based parameter-efficient variant for limited encoder adaptation. Experiments on both in-distribution and out-of-distribution benchmarks show that RobustMedSAM improves degraded-image Dice from 0.613 to 0.719 (+0.106) over SAM, demonstrating that structured fusion of complementary pretrained models is an effective and practical approach for robust medical image segmentation.