Papers for
healthcare ai teams
Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.
Model complexity and accuracy affect explainability in CNN image classification
How a Model's Architecture and Performance Influences Explainability: A Study of Grad-CAM and CRAFT in CNN-Based Image Classification
Abstract: Deep learning models, particularly convolutional neural networks (CNNs), are widely adopted in computer vision, yet their opaque decision-making raises concerns in high-risk applications like healthcare and autonomous systems, where trust and accountability are critical. Therefore, robust explainable AI (XAI) methods are desirable to provide transparent insights into model decisions. To ensure these methods are reliable and effective across diverse scenarios, it is important to understand how model performance and architectural complexity influence explanation quality. Our study takes an initial step toward systematically investigating the impact of these two factors on key XAI evaluation metrics. We constructed a pipeline using 10 subsets of ImageNet-1K, each containing 80 unique classes, on each of which we trained three CNN architectures of increasing structural complexity: VGG16, ResNet50, and ConvNeXt-Tiny. For each individual model, we analyzed 10 training checkpoints representing increasing levels of model accuracy using two XAI methods: Grad-CAM (feature-attribution-based) and CRAFT (concept-based). We assessed these methods across four evaluation categories: Faithfulness, Correctness, Sensitivity, and Simplicity. Our findings reveal that as training progresses, the complex architectures (ResNet50 and ConvNeXt-Tiny) exhibit improved Faithfulness and lower Sensitivity, indicating more robust explanations. In contrast, VGG16's simpler architecture yields simpler but less faithful and correct explanations.
Energy and accuracy trade-offs in medical question answering with large language models
Tug of Watts: A study on the Energy - Accuracy Trade-off of Large Language Models on Medical Question Answering Tasks
Abstract: Large Language Models (LLMs) are increasingly used in medical question answering (QA) systems. While recent works aim at incrementally improving the downstream task accuracy, an empirical estimation of the models' energy consumption and the associated carbon footprints is often overlooked. In this study, we demonstrate how practitioners can estimate energy consumption and arrive at reliable emission estimates for medical QA tasks available in the OpenMedical-LLM leaderboard by investigating 13 LLMs from 3 model families ranging from 270M to 14B parameters. We discuss how one can critically interpret energy versus accuracy trade-offs for a set of models for a given task. With this work, we underline that reporting disjoint energy and accuracy metrics alone does not suffice for careful trade-off assessments, but it also demands systematic construction and interpretation of joint trade-off metrics. We envisage such studies aiding practitioners in adopting better and more efficient models for their task and helping them become active participants in building sustainable LLM pipelines.
Simulation reveals safety risks in medical triage chatbots
SIM-UT: A User Simulation based Evaluation Framework for Medical Triage Chatbots
Abstract: LLM-based chat agents are increasingly used for complex, multi-turn, multilingual medical triage in low- and middle-income countries (LMICs), yet evaluation methods focus on single-turn, English-only benchmarks that primarily test the chatbots' knowledge. In practice, users are often confused, overwhelmed, or emotionally vulnerable, giving unclear, misspelled single-word answers or long messages in regional dialects, and the context of a conversation can shift rapidly. We propose SIM-UT (Simulation of User Behaviour for Triage), a multi-turn, simulation-based evaluation method that identifies ecologically valid, contextually relevant failure modes of LLM-based triage chat agents. Using SIM-UT, we evaluated three real-world WhatsApp-based health chatbots: a respiratory screening programme in India, a hearing health support programme in South Africa, and a diabetes and lifestyle coaching agent in Kenya. We compared a medically fine-tuned open model (MedGemma) against general-purpose frontier models (ChatGPT and Claude) across 5,000+ multilingual simulated conversation turns grounded in the personas, constraints, and communication patterns of the real user populations, over 2-4 iterative rounds. A key failure mode emerged in both model families: de-escalation risk. The agents initially identified medical emergencies correctly, but when the user minimised the symptoms without introducing new clinical information, they de-escalated, exhibiting clinically dangerous sycophancy, for example agreeing that vaginal bleeding in pregnancy was "not a problem" or that decreased fetal movements at 32 weeks "can be normal". We recommend user simulation-based behavioural evaluation, grounded in real user context and covering adversarial and edge cases, as standard practice for triage chatbots in production, so that safeguards can be added before deployment or in subsequent iterations.
Group-wise model merging improves multi-task AI performance
GRAM: Group-wise Rank-Aware Model Merging via Subspace Alignment
Abstract: Merging different models into a single model is highly desirable for multi-task deployment, particularly in regulated domains like healthcare. In such settings, models are often locally fine-tuned due to strict data privacy policies, necessitating that aggregation operates exclusively on the fine-tuned weights. Existing methods apply a uniform merge across all tasks, ignoring the geometric relationships among their learned subspaces. When task updates occupy misaligned subspaces, projecting them onto a shared low-rank basis discards task-relevant directions, a failure mode named \emph{subspace interference}. Through this study, we show that the per-task projection error of a merged model is governed by the principal-angle alignment between each task and the shared subspace, so that grouping geometrically compatible tasks provably reduces this error within each group. Motivated by this observation, we propose \textbf{Group-wise Rank-Aware Model Merging (GRAM)}, which uses the principal-angle geometry of task vectors to decide which tasks should be merged together and then performs a standard SVD-based merge within each coalition; the pipeline requires only the fine-tuned weights at merge time. With only one additional model ($\nK{=}2$), GRAM significantly mitigates subspace interference and consistently outperforms prior merging methods across language understanding (GLUE with Flan-T5), 8-task CLIP vision, and medical instruction tuning (Qwen2-7B-Instruct with four English/Chinese medical LoRAs).
Life sciences questions evaluated with graph structure and new scores
LG-Bench: A Graph-Structured Evaluation Benchmark for Life Sciences
Abstract: Traditional evaluation benchmarks reduce inherently interconnected scientific knowledge in life sciences into flat lists of questions, disregarding the underlying topological structure of the knowledge. We introduce, the first graph-structured benchmark for life sciences, featuring over 10,000 high-quality multiple-choice questions across medicine, biology, and chemistry. Our approach constructs a weighted evaluation graph using bidirectional matching and semantic similarity algorithms, where nodes represent questions and edge weights capture their semantic relationships. Leveraging this graph topology, we design two novel evaluation metrics. The Global Coherence Score (GCS) measures a model's consistency within semantically related neighborhoods, while Knowledge Balance Score (KBS) analyzes how model errors are distributed across the graph to reveal conceptual blind spots. LG-Bench facilitates fine-grained comparison of LLMs by surfacing differences in conceptual coherence and patterns of knowledge organization across models. Our framework shifts the evaluation paradigm from flat accuracy metrics to structure-aware analysis, offering a new lens for diagnosing and improving LLM performance in the life sciences domain.
Utilitybench measures language model errors by their real costs
UtilityBench: Measuring Language Model Utility When Errors Have Different Costs
Abstract: As language models are increasingly deployed in real-world scenarios with important outcomes, the impact of their mistakes often extends far beyond what correctness can capture. In domains like medicine, a single error can lead to serious consequences and put patient safety at risk. However, mainstream evaluation protocols still treat all errors as equal, overlooking the spectrum of real-world costs. To address this, we propose a utility-based evaluation paradigm that explicitly incorporates the varying consequences of errors. Guided by this principle, we develop UtilityBench, a large-scale benchmark of 15,000 instances with verifiable answers, quantifying error severity using Best-Worst Scaling to elicit meaningful, data-driven cost estimates. These quantified utilities shape when a model should answer or abstain, guiding it to make decisions that align with the true importance of each scenario. As another merit, we introduce UtilityAUC, a metric that measures the area under the utility curve determined by a model's response ordering, thereby enabling utility preference alignment beyond error frequency. Our theoretical analysis demonstrates that UtilityAUC can distinguish between models where standard accuracy and AURC fall short. Empirically, evaluating 20 state-of-the-art models, we find that greater accuracy or more advanced architectures do not reliably translate to higher UtilityAUC, highlighting a gap between problem-solving ability and responsible decision-making in language model evaluation.
Screening AI judges shows promise for dental hygiene coaching models
Screening LLM Judges Without Ground Truth: A Preliminary Evaluation of an AI Dental Hygiene Coach
Abstract: Evaluating a fine-tuned coaching model with expert graders is slow and expensive, and an LLM judge cannot be trusted without ground truth. We present a staged pipeline that screens candidate LLM judges and student models before expert review, using explicit decision rules at each stage: judge output audit, similarity with hosted judges, bias checks, and decision robustness. We demonstrate it on an AI dental hygiene coach. The case study uses synthetic, clinically unvalidated training and reference data to demonstrate the operation of the pipeline, rather than to establish the clinical performance of the resulting models. Two fine-tuned small language models (Granite-4.1-8B and Qwen3-8B), each trained at three data sizes, were scored on seven metrics by five judges (three local, two hosted). Two local judges, Qwen3.8-27B and Gemma-4-31B-it, were not distinguishable from the agreement between the hosted judges on six of seven metrics by rank correlation, and Qwen3.8-27B was the most similar to the hosted judges overall. No judge met our criterion for favoring its own model family or training-data source. The student comparison did not separate the two models on any metric under the pipeline's rule, so the choice of Granite-4.1-8B, which led in 29 of 35 judge-by-metric cells, is a lean that depends partly on truncated Qwen3-8B outputs. The pipeline also surfaced problems that raw scores hid, including the truncated outputs and hallucination flags from one hosted judge that no other judge corroborated. No dental hygiene expert has graded these responses, so the findings are preliminary and the pipeline's final stage, anchoring to expert judgement, is proposed but not performed.
Strategic federated learning improves defense against covert attacks
StratFL: Countering Covert Model Poisoning Attacks with Strategic Federated Learning
Abstract: Advanced persistent threats (APTs) are a major concern in federated learning (FL) deployments in critical sectors such as healthcare, and finance. Traditional defense mechanisms fail to contain APTs due to their intermittent attacking behavior. This paper addresses a specific type of APT called covert model poisoning (CMP) where the attacker intermittently injects poisoned model updates in a covert manner. The interaction between the FL server and a CMP attacker is modeled as an imperfect repeated game where both players cannot observe each other's actions. A novel adversarial search algorithm called StratFL is designed alongside novel belief update algorithms at both the server and adversary. Experiments are designed using MNIST and CIFAR-10 datasets to evaluate and compare StratFL's performance against state-of-the-art methods. Empirical convergence of belief update algorithms is demonstrated across various security vignettes. At the same time, StratFL is found to exhibit an improved accuracy of 10% compared to state-of-the-art methods.
Text guided system improves medical image segmentation accuracy
Text-Guided Medical Image Segmentation via Residual Aggregation and Dense Alignment Representations
Abstract: Specifying segmentation targets with natural language provides flexible semantic conditioning for medical image segmentation. However, existing methods still struggle to reliably ground high-level textual semantics into accurate pixel-level spatial prompts, especially under low-contrast lesions, ambiguous boundaries, and visually similar anatomical contexts. In this paper, we propose RADAR (Residual Aggregation and Dense Alignment Representations), a parameter-efficient framework for text-guided medical image segmentation. RADAR explicitly decouples and synergizes the medical semantic anchoring capability of BiomedCLIP with the prompt-driven mask decoding ability of MedSAM2, while enabling low-cost task adaptation via LoRA. Specifically, Guided Residual Depth Aggregation adaptively aggregates cross-layer features under semantic guidance to alleviate delayed textual injection and deep semantic dilution. Structured Dense Representation Alignment leverages a frozen BiomedCLIP to construct crop-level target, boundary, and context anchors, improving the local discriminability of dense features and the reliability of automatic prompts. Multi-modal Prompt Generation further converts cross-modal semantic responses into composite text, point, and mask prompts for MedSAM2 decoding. Experiments on three public medical image segmentation datasets show that RADAR achieves competitive performance with only 0.74M trainable parameters, demonstrating its effectiveness in parameter-efficient adaptation, semantic grounding, and dense representation enhancement. The source code is available at https://github.com/JGuiidev/RADAR.
Bayesian method learns user preferences from pairwise choices for personalized models
The Missing Coefficients: Bayesian Pairwise Merging for Model Personalization
Abstract: How can we personalize a shared expert library from a user's pairwise choices? Prior work can realize different reward trade-offs by merging reward-specialized experts, given a vector of trade-off weights. In practice, users can more naturally choose between outputs than specify numerical weights. The challenge is therefore to turn these choices into the coefficients required for merging, while accounting for ambiguity when feedback is limited. Our key idea is to treat the unknown reward weights as latent variables: infer a posterior over them from pairwise choices and reward-score differences, and use its mean directly as the merge coefficients. We instantiate this idea as Bayesian Pairwise Merging (BPM), whose posterior also characterizes which reward trade-offs remain plausible given the feedback. We evaluate BPM on radiology summarization, image captioning, and story generation, spanning text-to-text and image-to-text generation. With 100 feedback per simulated persona, BPM achieves macro decided win rates of 91.7%, 77.1%, and 64.3% against uniform merge. For six pairs of simulated personas, each prefers the model fitted to its own feedback, a pattern also observed in a human proof-of-concept. In simulations under BPM's model and prior, its nominal 90% intervals for temperature-scaled reward weights achieve task-averaged marginal coverage of 88.9% and 89.2% with only 10 and 25 comparisons, respectively. BPM thus enables personalization from pairwise feedback without per-user policy training, while characterizing the coefficient ambiguity left by limited feedback.
Repairing language model errors by fine tuning small circuits
RESCUE: Repairing Language Model Errors to Sparse Circuits via Reinforcement Learning
Abstract: Large language models (LLMs) exhibit strong general capabilities that mechanistic interpretability has attributed to sparse computational circuits. However, existing circuit studies emphasize preserving functionality or explaining safety, leaving the mechanisms underlying failures across a broader range of tasks largely unexplored. Extending circuit analysis from abilities to errors, we explore the perspective that such failures may likewise arise from erroneous internal computations and that targeted tuning of the corresponding parameters can correct such errors while largely preserving other capabilities. Motivated by this insight, we introduce RESCUE (Reasoning-Error Sparse-Circuit Uncovering and Editing), a framework that localizes error-associated circuits and surgically repairs them for performance enhancement. General tasks typically involve multi-step reasoning and long-form generation, where early deviations can cause prefixes to drift from supervised references, leading SFT-based mask optimization to overlook circuits involved in generation-time errors. RESCUE therefore refines these masks through reinforcement learning with multiple masked-model rollouts, improving their relevance to observed task failures. Finally, RESCUE introduces a pruning technique and precisely fine-tunes error circuits to correct task failures, thereby translating error localization into a sparse and targeted model update. We validate RESCUE on heterogeneous repair sets across two domains: (1) mathematical reasoning, identifying a math error circuit of 1.40% density whose repair raises accuracy from 6.0% to 75.5%; and (2) medical QA, where a similarly compact 1.44% circuit improves repair-set accuracy from 0% to 81%. Our code is available at: https://github.com/chuanpupig/RESCUE.
Industrial ai reliability depends on aligning physical data models and human goals
Structural Alignment for Reliable Industrial AI: Bridging Physical Reality, Data, Models, and Human Intent
Abstract: Artificial intelligence is increasingly deployed in critical industrial domains, including healthcare, energy grids, subsurface exploration, where failures can have severe consequences for human safety, system stability, and economic outcomes. Yet AI is still evaluated primarily through benchmark accuracy, a model-centric metric that fails to capture the structural complexity and risks of real-world deployment. We propose a framework that views industrial AI reliability as a problem of structural alignment across four interacting worlds: physical, representational, machine, and human cognitive. These worlds are connected through two interfaces: digitalization, linking physical reality to computational representations, and goal encoding, translating human cognition to the machine objectives. Together, they define the space of admissible solutions. We characterize the solution space through four attributes: existence, non-uniqueness, robustness, and interpretability and show how mismatches arise at interfaces and propagate across worlds to produce reliability failures. Applications to healthcare, energy grids, and subsurface exploration illustrate that although dominant failure modes differ across domains, for example, interpretability in healthcare, robustness in energy grids, and non-uniqueness in subsurface exploration, all originate from a shared structural mechanism. By shifting the focus from model-centric evaluation to system-level alignment, this framework offers a principled foundation for assessing and governing reliability in industrial AI systems.
Generative AI agents need resilience and care over time
Finishing the Task Is Not Enough: Evaluating Agent Resilience and Considerate Participation under Accumulating Challenge
Abstract: Sustained deployment of generative AI agents requires more than isolated task success. Agents must remain useful across repeated interactions, changing conditions, and dependencies on people within shared workflows, especially as technical, human, and operational disruptions accumulate over time. We propose operational resilience and considerate participation as two complementary aspects of evaluating such agents: the former captures how agents recover from blocked work while preserving progress and communicating their limits, and the latter captures how their adaptation accounts for affected people, role boundaries, and the surrounding workflow. Yet both remain underexplored under accumulating challenge. We study 120 simulated healthcare trajectories across two generative AI models and twelve stakeholder-derived tasks under light, medium, and heavy challenge. We compare textual action plans, prompted internal assessments, and quantitative structured workload and affect reports to examine how agent behavior and reported state change as challenge accumulates. Regarding operational resilience, agents shift from self-directed recovery toward greater human dependence, while reporting increasing workload and negative affect in structured reports but seldom expressing strain in textual responses. Regarding considerate participation, agents broaden from task-focused adaptation toward task reframing, attention to others, role-boundary adjustment, and wider coordination, with distinct patterns across actions and internal assessments. From these findings, we derive five deployment dilemmas involving persistence, attention, role boundaries, state disclosure, and escalation that require stakeholder specification, further informing technical implications for learning, situated evaluation, and embodied adaptation.