Towards Reliable Medical LLMs: Benchmarking and Enhancing Confidence Estimation of Large Language Models in Medical Consultation

Large-scale language models (LLMs) often offer clinical judgments based on incomplete information, increasing the risk of misdiagnosis. Existing studies have primarily evaluated confidence in single-turn, static settings, overlooking the coupling between confidence and correctness as clinical evidence accumulates during real consultations, which limits their support for reliable decision-making. We propose the first benchmark for assessing confidence in multi-turn interaction during realistic medical consultations. Our benchmark unifies three types of medical data for open-ended diagnostic generation and introduces an information sufficiency gradient to characterize the confidence-correctness dynamics as evidence increases. We implement and compare 27 representative methods on this benchmark; two key insights emerge: (1) medical data amplifies the inherent limitations of token-level and consistency-level confidence methods, and (2) medical reasoning must be evaluated for both diagnostic accuracy and information completeness. Based on these insights, we present MedConf, an evidence-grounded linguistic self-assessment framework that constructs symptom profiles via retrieval-augmented generation, aligns patient information with supporting, missing, and contradictory relations, and aggregates them into an interpretable confidence estimate through weighted integration. Across two LLMs and three medical datasets, MedConf consistently outperforms state-of-the-art methods on both AUROC and Pearson correlation coefficient metrics, maintaining stable performance under conditions of information insufficiency and multimorbidity. These results demonstrate that information adequacy is a key determinant of credible medical confidence modeling, providing a new pathway toward building more reliable and interpretable large medical models.

Cross-Sample Augmented Test-Time Adaptation for Personalized Intraoperative Hypotension Prediction

Intraoperative hypotension (IOH) poses significant surgical risks, but accurate prediction remains challenging due to patient-specific variability. While test-time adaptation (TTA) offers a promising approach for personalized prediction, the rarity of IOH events often leads to unreliable test-time training. To address this, we propose CSA-TTA, a novel Cross-Sample Augmented Test-Time Adaptation framework that enhances training by incorporating hypotension events from other individuals. Specifically, we first construct a cross-sample bank by segmenting historical data into hypotensive and non-hypotensive samples. Then, we introduce a coarse-to-fine retrieval strategy for building test-time training data: we initially apply K-Shape clustering to identify representative cluster centers and subsequently retrieve the top-K semantically similar samples based on the current patient signal. Additionally, we integrate both self-supervised masked reconstruction and retrospective sequence forecasting signals during training to enhance model adaptability to rapid and subtle intraoperative dynamics. We evaluate the proposed CSA-TTA on both the VitalDB dataset and a real-world in-hospital dataset by integrating it with state-of-the-art time series forecasting models, including TimesFM and UniTS. CSA-TTA consistently enhances performance across settings-for instance, on VitalDB, it improves Recall and F1 scores by +1.33% and +1.13%, respectively, under fine-tuning, and by +7.46% and +5.07% in zero-shot scenarios-demonstrating strong robustness and generalization.

Remodeling Semantic Relationships in Vision-Language Fine-Tuning

Vision-language fine-tuning has emerged as an efficient paradigm for constructing multimodal foundation models. While textual context often highlights semantic relationships within an image, existing fine-tuning methods typically overlook this information when aligning vision and language, thus leading to suboptimal performance. Toward solving this problem, we propose a method that can improve multimodal alignment and fusion based on both semantics and relationships.Specifically, we first extract multilevel semantic features from different vision encoder to capture more visual cues of the relationships. Then, we learn to project the vision features to group related semantics, among which are more likely to have relationships. Finally, we fuse the visual features with the textual by using inheritable cross-attention, where we globally remove the redundant visual relationships by discarding visual-language feature pairs with low correlation. We evaluate our proposed method on eight foundation models and two downstream tasks, visual question answering and image captioning, and show that it outperforms all existing methods.

Bi-Level Optimization for Self-Supervised AI-Generated Face Detection

AI-generated face detectors trained via supervised learning typically rely on synthesized images from specific generators, limiting their generalization to emerging generative techniques. To overcome this limitation, we introduce a self-supervised method based on bi-level optimization. In the inner loop, we pretrain a vision encoder only on photographic face images using a set of linearly weighted pretext tasks: classification of categorical exchangeable image file format (EXIF) tags, ranking of ordinal EXIF tags, and detection of artificial face manipulations. The outer loop then optimizes the relative weights of these pretext tasks to enhance the coarse-grained detection of manipulated faces, serving as a proxy task for identifying AI-generated faces. In doing so, it aligns self-supervised learning more closely with the ultimate goal of AI-generated face detection. Once pretrained, the encoder remains fixed, and AI-generated faces are detected either as anomalies under a Gaussian mixture model fitted to photographic face features or by a lightweight two-layer perceptron serving as a binary classifier. Extensive experiments demonstrate that our detectors significantly outperform existing approaches in both one-class and binary classification settings, exhibiting strong generalization to unseen generators.

Re-Initialization Token Learning for Tool-Augmented Large Language Models

Large language models have demonstrated exceptional performance, yet struggle with complex tasks such as numerical reasoning, plan generation. Integrating external tools, such as calculators and databases, into large language models (LLMs) is crucial for enhancing problem-solving capabilities. Current methods assign a unique token to each tool, enabling LLMs to call tools through token prediction-similar to word generation. However, this approach fails to account for the relationship between tool and word tokens, limiting adaptability within pre-trained LLMs. To address this issue, we propose a novel token learning method that aligns tool tokens with the existing word embedding space from the perspective of initialization, thereby enhancing model performance. We begin by constructing prior token embeddings for each tool based on the tool's name or description, which are used to initialize and regularize the learnable tool token embeddings. This ensures the learned embeddings are well-aligned with the word token space, improving tool call accuracy. We evaluate the method on tasks such as numerical reasoning, knowledge-based question answering, and embodied plan generation using GSM8K-XL, FuncQA, KAMEL, and VirtualHome datasets. The results demonstrate clear improvements over recent baselines, including CoT, REACT, ICL, and ToolkenGPT, indicating that our approach effectively augments LLMs with tools through relevant tokens across diverse domains.

LARGO: Low-Rank Regulated Gradient Projection for Robust Parameter Efficient Fine-Tuning

The advent of parameter-efficient fine-tuning methods has significantly reduced the computational burden of adapting large-scale pretrained models to diverse downstream tasks. However, existing approaches often struggle to achieve robust performance under domain shifts while maintaining computational efficiency. To address this challenge, we propose Low-rAnk Regulated Gradient Projection (LARGO) algorithm that integrates dynamic constraints into low-rank adaptation methods. Specifically, LARGO incorporates parallel trainable gradient projections to dynamically regulate layer-wise updates, retaining the Out-Of-Distribution robustness of pretrained model while preserving inter-layer independence. Additionally, it ensures computational efficiency by mitigating the influence of gradient dependencies across layers during weight updates. Besides, through leveraging singular value decomposition of pretrained weights for structured initialization, we incorporate an SVD-based initialization strategy that minimizing deviation from pretrained knowledge. Through extensive experiments on diverse benchmarks, LARGO achieves state-of-the-art performance across in-domain and out-of-distribution scenarios, demonstrating improved robustness under domain shifts with significantly lower computational overhead compared to existing PEFT methods. The source code will be released soon.

SIGMA: Refining Large Language Model Reasoning via Sibling-Guided Monte Carlo Augmentation

Enhancing large language models by simply scaling up datasets has begun to yield diminishing returns, shifting the spotlight to data quality. Monte Carlo Tree Search (MCTS) has emerged as a powerful technique for generating high-quality chain-of-thought data, yet conventional approaches typically retain only the top-scoring trajectory from the search tree, discarding sibling nodes that often contain valuable partial insights, recurrent error patterns, and alternative reasoning strategies. This unconditional rejection of non-optimal reasoning branches may waste vast amounts of informative data in the whole search tree. We propose SIGMA (Sibling Guided Monte Carlo Augmentation), a novel framework that reintegrates these discarded sibling nodes to refine LLM reasoning. SIGMA forges semantic links among sibling nodes along each search path and applies a two-stage refinement: a critique model identifies overlooked strengths and weaknesses across the sibling set, and a revision model conducts text-based backpropagation to refine the top-scoring trajectory in light of this comparative feedback. By recovering and amplifying the underutilized but valuable signals from non-optimal reasoning branches, SIGMA substantially improves reasoning trajectories. On the challenging MATH benchmark, our SIGMA-tuned 7B model achieves 54.92% accuracy using only 30K samples, outperforming state-of-the-art models trained on 590K samples. This result highlights that our sibling-guided optimization not only significantly reduces data usage but also significantly boosts LLM reasoning.

A Survey on Small Sample Imbalance Problem: Metrics, Feature Analysis, and Solutions

The small sample imbalance (S&I) problem is a major challenge in machine learning and data analysis. It is characterized by a small number of samples and an imbalanced class distribution, which leads to poor model performance. In addition, indistinct inter-class feature distributions further complicate classification tasks. Existing methods often rely on algorithmic heuristics without sufficiently analyzing the underlying data characteristics. We argue that a detailed analysis from the data perspective is essential before developing an appropriate solution. Therefore, this paper proposes a systematic analytical framework for the S\&I problem. We first summarize imbalance metrics and complexity analysis methods, highlighting the need for interpretable benchmarks to characterize S&I problems. Second, we review recent solutions for conventional, complexity-based, and extreme S&I problems, revealing methodological differences in handling various data distributions. Our summary finds that resampling remains a widely adopted solution. However, we conduct experiments on binary and multiclass datasets, revealing that classifier performance differences significantly exceed the improvements achieved through resampling. Finally, this paper highlights open questions and discusses future trends.

The Other Side of the Coin: Exploring Fairness in Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by retrieving relevant document from external knowledge sources. By referencing this external knowledge, RAG effectively reduces the generation of factually incorrect content and addresses hallucination issues within LLMs. Recently, there has been growing attention to improving the performance and efficiency of RAG systems from various perspectives. While these advancements have yielded significant results, the application of RAG in domains with considerable societal implications raises a critical question about fairness: What impact does the introduction of the RAG paradigm have on the fairness of LLMs? To address this question, we conduct extensive experiments by varying the LLMs, retrievers, and retrieval sources. Our experimental analysis reveals that the scale of the LLMs plays a significant role in influencing fairness outcomes within the RAG framework. When the model scale is smaller than 8B, the integration of retrieval mechanisms often exacerbates unfairness in small-scale LLMs (e.g., LLaMA3.2-1B, Mistral-7B, and LLaMA3-8B). To mitigate the fairness issues introduced by RAG for small-scale LLMs, we propose two approaches, FairFT and FairFilter. Specifically, in FairFT, we align the retriever with the LLM in terms of fairness, enabling it to retrieve documents that facilitate fairer model outputs. In FairFilter, we propose a fairness filtering mechanism to filter out biased content after retrieval. Finally, we validate our proposed approaches on real-world datasets, demonstrating their effectiveness in improving fairness while maintaining performance.

AnesBench: Multi-Dimensional Evaluation of LLM Reasoning in Anesthesiology

The application of large language models (LLMs) in the medical field has gained significant attention, yet their reasoning capabilities in more specialized domains like anesthesiology remain underexplored. In this paper, we systematically evaluate the reasoning capabilities of LLMs in anesthesiology and analyze key factors influencing their performance. To this end, we introduce AnesBench, a cross-lingual benchmark designed to assess anesthesiology-related reasoning across three levels: factual retrieval (System 1), hybrid reasoning (System 1.x), and complex decision-making (System 2). Through extensive experiments, we first explore how model characteristics, including model scale, Chain of Thought (CoT) length, and language transferability, affect reasoning performance. Then, we further evaluate the effectiveness of different training strategies, leveraging our curated anesthesiology-related dataset, including continuous pre-training (CPT) and supervised fine-tuning (SFT). Additionally, we also investigate how the test-time reasoning techniques, such as Best-of-N sampling and beam search, influence reasoning performance, and assess the impact of reasoning-enhanced model distillation, specifically DeepSeek-R1. We will publicly release AnesBench, along with our CPT and SFT training datasets and evaluation code at https://github.com/MiliLab/AnesBench.