Aidong Zhang

Guangzhi Xiong, Qiao Jin, Xiao Wang et al.

The emergent abilities of large language models (LLMs) have demonstrated great potential in solving medical questions. They can possess considerable medical knowledge, but may still hallucinate and are inflexible in the knowledge updates. While Retrieval-Augmented Generation (RAG) has been proposed to enhance the medical question-answering capabilities of LLMs with external knowledge bases, it may still fail in complex cases where multiple rounds of information-seeking are required. To address such an issue, we propose iterative RAG for medicine (i-MedRAG), where LLMs can iteratively ask follow-up queries based on previous information-seeking attempts. In each iteration of i-MedRAG, the follow-up queries will be answered by a conventional RAG system and they will be further used to guide the query generation in the next iteration. Our experiments show the improved performance of various LLMs brought by i-MedRAG compared with conventional RAG on complex questions from clinical vignettes in the United States Medical Licensing Examination (USMLE), as well as various knowledge tests in the Massive Multitask Language Understanding (MMLU) dataset. Notably, our zero-shot i-MedRAG outperforms all existing prompt engineering and fine-tuning methods on GPT-3.5, achieving an accuracy of 69.68% on the MedQA dataset. In addition, we characterize the scaling properties of i-MedRAG with different iterations of follow-up queries and different numbers of queries per iteration. Our case studies show that i-MedRAG can flexibly ask follow-up queries to form reasoning chains, providing an in-depth analysis of medical questions. To the best of our knowledge, this is the first-of-its-kind study on incorporating follow-up queries into medical RAG. The implementation of i-MedRAG is available at https://github.com/Teddy-XiongGZ/MedRAG.

Guangtao Zheng, Wenqian Ye, Aidong Zhang

Few-shot image classifiers are designed to recognize and classify new data with minimal supervision and limited data but often show reliance on spurious correlations between classes and spurious attributes, known as spurious bias. Spurious correlations commonly hold in certain samples and few-shot classifiers can suffer from spurious bias induced from them. There is an absence of an automatic benchmarking system to assess the robustness of few-shot classifiers against spurious bias. In this paper, we propose a systematic and rigorous benchmark framework, termed FewSTAB, to fairly demonstrate and quantify varied degrees of robustness of few-shot classifiers to spurious bias. FewSTAB creates few-shot evaluation tasks with biased attributes so that using them for predictions can demonstrate poor performance. To construct these tasks, we propose attribute-based sample selection strategies based on a pre-trained vision-language model, eliminating the need for manual dataset curation. This allows FewSTAB to automatically benchmark spurious bias using any existing test data. FewSTAB offers evaluation results in a new dimension along with a new design guideline for building robust classifiers. Moreover, it can benchmark spurious bias in varied degrees and enable designs for varied degrees of robustness. Its effectiveness is demonstrated through experiments on ten few-shot learning methods across three datasets. We hope our framework can inspire new designs of robust few-shot classifiers. Our code is available at https://github.com/gtzheng/FewSTAB.

Sanchit Sinha, Guangzhi Xiong, Bohan Liu et al.

The effectiveness of Chain-of-Thought (CoT) prompting in Multimodal Large Language Models (MLLMs) remains uncertain: across several visual reasoning benchmarks, CoT prompting often degrades performance compared to direct prompting. In this paper, we provide a systematic analysis of CoT behavior in three modern MLLM families across model scales on datasets requiring step-wise visual evidence. Our analysis identifies two recurring failure modes: premature answer commitment and limited direct visual-token access during rationale generation. We further find that standard CoT-style Supervised Fine-Tuning (CoT-SFT) can mitigate these issues only partially, while often increasing reliance on textual priors and reducing counterfactual visual dependence. Motivated by these findings, we propose Attentive-CoT (Att-CoT), an attention-guided fine-tuning objective that encourages CoT trajectories to delay answer commitment while maintaining sustained visual-token access. Att-CoT can be plugged into any CoT-SFT training run without architectural changes. Experiments on three visual reasoning benchmarks across six MLLMs show that Att-CoT enhances CoT performance over standard fine-tuning.

Jing Ma, Ruocheng Guo, Saumitra Mishra et al.

Counterfactual explanations promote explainability in machine learning models by answering the question "how should an input instance be perturbed to obtain a desired predicted label?". The comparison of this instance before and after perturbation can enhance human interpretation. Most existing studies on counterfactual explanations are limited in tabular data or image data. In this work, we study the problem of counterfactual explanation generation on graphs. A few studies have explored counterfactual explanations on graphs, but many challenges of this problem are still not well-addressed: 1) optimizing in the discrete and disorganized space of graphs; 2) generalizing on unseen graphs; and 3) maintaining the causality in the generated counterfactuals without prior knowledge of the causal model. To tackle these challenges, we propose a novel framework CLEAR which aims to generate counterfactual explanations on graphs for graph-level prediction models. Specifically, CLEAR leverages a graph variational autoencoder based mechanism to facilitate its optimization and generalization, and promotes causality by leveraging an auxiliary variable to better identify the underlying causal model. Extensive experiments on both synthetic and real-world graphs validate the superiority of CLEAR over the state-of-the-art methods in different aspects.

Sanchit Sinha, Mengdi Huai, Jianhui Sun et al.

Rising usage of deep neural networks to perform decision making in critical applications like medical diagnosis and financial analysis have raised concerns regarding their reliability and trustworthiness. As automated systems become more mainstream, it is important their decisions be transparent, reliable and understandable by humans for better trust and confidence. To this effect, concept-based models such as Concept Bottleneck Models (CBMs) and Self-Explaining Neural Networks (SENN) have been proposed which constrain the latent space of a model to represent high level concepts easily understood by domain experts in the field. Although concept-based models promise a good approach to both increasing explainability and reliability, it is yet to be shown if they demonstrate robustness and output consistent concepts under systematic perturbations to their inputs. To better understand performance of concept-based models on curated malicious samples, in this paper, we aim to study their robustness to adversarial perturbations, which are also known as the imperceptible changes to the input data that are crafted by an attacker to fool a well-learned concept-based model. Specifically, we first propose and analyze different malicious attacks to evaluate the security vulnerability of concept based models. Subsequently, we propose a potential general adversarial training-based defense mechanism to increase robustness of these systems to the proposed malicious attacks. Extensive experiments on one synthetic and two real-world datasets demonstrate the effectiveness of the proposed attacks and the defense approach.

Guangzhi Xiong, Qiao Jin, Sanchit Sinha et al.

Large Vision Language Models (LVLMs) show promise in medical applications, but their inability to faithfully ground responses in visual evidence raises serious concerns about clinical trustworthiness. While visual attribution methods are widely used to explain LVLM predictions, whether these explanations actually reflect the visual evidence underlying the model's decision is largely unverified, since ground-truth annotations for internal model reasoning are typically unavailable. We address this question for chest X-ray (CXR) reasoning by developing a causal evaluation framework that retains only CXR-VQA samples for which the expert-annotated region is verified, via counterfactual editing, to be causally responsible for the model's prediction. Using this framework across 11 attribution methods, six open-source LVLMs, and two output modes (direct answer and step-by-step reasoning), we find that existing attribution methods often fail to identify the evidence used by LVLMs. To address this failure, we propose MedFocus, a concept-based attribution method that localizes clinically meaningful anatomical regions via unbalanced optimal transport and measures their causal effect on model outputs through targeted interventions. MedFocus produces spatial, concept-level, and token-level attributions and substantially outperforms prior methods, taking a step toward more trustworthy attribution for medical LVLMs. Our data and code are available at https://github.com/gzxiong/medfocus/.

Jing Ma, Ruocheng Guo, Aidong Zhang et al.

Fairness-aware machine learning has attracted a surge of attention in many domains, such as online advertising, personalized recommendation, and social media analysis in web applications. Fairness-aware machine learning aims to eliminate biases of learning models against certain subgroups described by certain protected (sensitive) attributes such as race, gender, and age. Among many existing fairness notions, counterfactual fairness is a popular notion defined from a causal perspective. It measures the fairness of a predictor by comparing the prediction of each individual in the original world and that in the counterfactual worlds in which the value of the sensitive attribute is modified. A prerequisite for existing methods to achieve counterfactual fairness is the prior human knowledge of the causal model for the data. However, in real-world scenarios, the underlying causal model is often unknown, and acquiring such human knowledge could be very difficult. In these scenarios, it is risky to directly trust the causal models obtained from information sources with unknown reliability and even causal discovery methods, as incorrect causal models can consequently bring biases to the predictor and lead to unfair predictions. In this work, we address the problem of counterfactually fair prediction from observational data without given causal models by proposing a novel framework CLAIRE. Specifically, under certain general assumptions, CLAIRE effectively mitigates the biases from the sensitive attribute with a representation learning framework based on counterfactual data augmentation and an invariant penalty. Experiments conducted on both synthetic and real-world datasets validate the superiority of CLAIRE in both counterfactual fairness and prediction performance.

Guangzhi Xiong, Sanchit Sinha, Zhenghao He et al.

Vision-language models (VLMs) have achieved impressive performance across a wide range of multimodal reasoning tasks, but they often struggle to disentangle fine-grained visual attributes and reason about underlying causal relationships. In-context learning (ICL) offers a promising avenue for VLMs to adapt to new tasks, but its effectiveness critically depends on the selection of demonstration examples. Existing retrieval-augmented approaches typically rely on passive similarity-based retrieval, which tends to select correlated but non-causal examples, amplifying spurious associations and limiting model robustness. We introduce CIRCLES (Composed Image Retrieval for Causal Learning Example Selection), a novel framework that actively constructs demonstration sets by retrieving counterfactual-style examples through targeted, attribute-guided composed image retrieval. By incorporating counterfactual-style examples, CIRCLES enables VLMs to implicitly reason about the causal relations between attributes and outcomes, moving beyond superficial correlations and fostering more robust and grounded reasoning. Comprehensive experiments on four diverse datasets demonstrate that CIRCLES consistently outperforms existing methods across multiple architectures, especially on small-scale models, with pronounced gains under information scarcity. Furthermore, CIRCLES retrieves more diverse and causally informative examples, providing qualitative insights into how models leverage in-context demonstrations for improved reasoning. Our code is available at https://github.com/gzxiong/CIRCLES.

Guangzhi Xiong, Sanchit Sinha, Aidong Zhang

The trade-off between interpretability and accuracy remains a core challenge in machine learning. Standard Generalized Additive Models (GAMs) offer clear feature attributions but are often constrained by their strictly additive nature, which can limit predictive performance. Introducing feature interactions can boost accuracy yet may obscure individual feature contributions. To address these issues, we propose Neural Additive Experts (NAEs), a novel framework that seamlessly balances interpretability and accuracy. NAEs employ a mixture of experts framework, learning multiple specialized networks per feature, while a dynamic gating mechanism integrates information across features, thereby relaxing rigid additive constraints. Furthermore, we propose targeted regularization techniques to mitigate variance among expert predictions, facilitating a smooth transition from an exclusively additive model to one that captures intricate feature interactions while maintaining clarity in feature attributions. Our theoretical analysis and experiments on synthetic data illustrate the model's flexibility, and extensive evaluations on real-world datasets confirm that NAEs achieve an optimal balance between predictive accuracy and transparent, feature-level explanations. The code is available at https://github.com/Teddy-XiongGZ/NAE.

Guangzhi Xiong, Zhenghao He, Bohan Liu et al.

Retrieval-Augmented Generation (RAG) improves the factuality of large language models (LLMs) by grounding outputs in retrieved evidence, but faithfulness failures, where generations contradict or extend beyond the provided sources, remain a critical challenge. Existing hallucination detection methods for RAG often rely either on large-scale detector training, which requires substantial annotated data, or on querying external LLM judges, which leads to high inference costs. Although some approaches attempt to leverage internal representations of LLMs for hallucination detection, their accuracy remains limited. Motivated by recent advances in mechanistic interpretability, we employ sparse autoencoders (SAEs) to disentangle internal activations, successfully identifying features that are specifically triggered during RAG hallucinations. Building on a systematic pipeline of information-based feature selection and additive feature modeling, we introduce RAGLens, a lightweight hallucination detector that accurately flags unfaithful RAG outputs using LLM internal representations. RAGLens not only achieves superior detection performance compared to existing methods, but also provides interpretable rationales for its decisions, enabling effective post-hoc mitigation of unfaithful RAG. Finally, we justify our design choices and reveal new insights into the distribution of hallucination-related signals within LLMs. The code is available at https://github.com/Teddy-XiongGZ/RAGLens.

Jiayi Chen, Hanjun Dai, Bo Dai et al.

Visually-rich document entity retrieval (VDER), which extracts key information (e.g. date, address) from document images like invoices and receipts, has become an important topic in industrial NLP applications. The emergence of new document types at a constant pace, each with its unique entity types, presents a unique challenge: many documents contain unseen entity types that occur only a couple of times. Addressing this challenge requires models to have the ability of learning entities in a few-shot manner. However, prior works for Few-shot VDER mainly address the problem at the document level with a predefined global entity space, which doesn't account for the entity-level few-shot scenario: target entity types are locally personalized by each task and entity occurrences vary significantly among documents. To address this unexplored scenario, this paper studies a novel entity-level few-shot VDER task. The challenges lie in the uniqueness of the label space for each task and the increased complexity of out-of-distribution (OOD) contents. To tackle this novel task, we present a task-aware meta-learning based framework, with a central focus on achieving effective task personalization that distinguishes between in-task and out-of-task distribution. Specifically, we adopt a hierarchical decoder (HC) and employ contrastive learning (ContrastProtoNet) to achieve this goal. Furthermore, we introduce a new dataset, FewVEX, to boost future research in the field of entity-level few-shot VDER. Experimental results demonstrate our approaches significantly improve the robustness of popular meta-learning baselines.

Xidong Wu, Jianhui Sun, Zhengmian Hu et al.

The minimax problems arise throughout machine learning applications, ranging from adversarial training and policy evaluation in reinforcement learning to AUROC maximization. To address the large-scale data challenges across multiple clients with communication-efficient distributed training, federated learning (FL) is gaining popularity. Many optimization algorithms for minimax problems have been developed in the centralized setting (\emph{i.e.} single-machine). Nonetheless, the algorithm for minimax problems under FL is still underexplored. In this paper, we study a class of federated nonconvex minimax optimization problems. We propose FL algorithms (FedSGDA+ and FedSGDA-M) and reduce existing complexity results for the most common minimax problems. For nonconvex-concave problems, we propose FedSGDA+ and reduce the communication complexity to $O(\varepsilon^{-6})$. Under nonconvex-strongly-concave and nonconvex-PL minimax settings, we prove that FedSGDA-M has the best-known sample complexity of $O(κ^{3} N^{-1}\varepsilon^{-3})$ and the best-known communication complexity of $O(κ^{2}\varepsilon^{-2})$. FedSGDA-M is the first algorithm to match the best sample complexity $O(\varepsilon^{-3})$ achieved by the single-machine method under the nonconvex-strongly-concave setting. Extensive experimental results on fair classification and AUROC maximization show the efficiency of our algorithms.

Wenqian Ye, Bo Yuan, Zhichao Xu et al.

We study a class of emergent misalignment in multi-agent systems (MAS), with a focus on automated workflows, which we refer to agentic misalignment. Although these systems can solve complex tasks, they often fail because agents act according to implicit proxy utilities that do not align with the intended human goals. We formally define these behaviors and analyze them within a Bayesian framework, showing that generic utilities naturally lead to posterior collapse of agents in automated workflows. To address this issue, we propose Agentic Evidence Attribution (AEA), a novel alignment paradigm that improves agent posteriors using context-specific evidence. AEA reasons over agent actions and provides structured evidence to correct misaligned behavior during collaboration. To better understand the role of evidence, we study two instantiations of AEA: self-reflection (internal evidence from the model) and weak-to-strong generalization (external evidence on the agentic trajectory). We show that a small evidence model effectively aligns the MAS by providing orthogonal failure attribution. Our results clarify the sources of agentic misalignment in automated workflows and show that evidence-based alignment can effectively improve agent collaboration and leads to reliable multi-agent systems built on automated workflows.

Xidong Wu, Jianhui Sun, Zhengmian Hu et al.

Conditional stochastic optimization has found applications in a wide range of machine learning tasks, such as invariant learning, AUPRC maximization, and meta-learning. As the demand for training models with large-scale distributed data grows in these applications, there is an increasing need for communication-efficient distributed optimization algorithms, such as federated learning algorithms. This paper considers the nonconvex conditional stochastic optimization in federated learning and proposes the first federated conditional stochastic optimization algorithm (FCSG) with a conditional stochastic gradient estimator and a momentum-based algorithm (FCSG-M). To match the lower bound complexity in the single-machine setting, we design an accelerated algorithm (Acc-FCSG-M) via the variance reduction to achieve the best sample and communication complexity. Compared with the existing optimization analysis for MAML in FL, federated conditional stochastic optimization considers the sample of tasks. Extensive experimental results on various tasks validate the efficiency of these algorithms.

Guangtao Zheng, Mengdi Huai, Aidong Zhang

Single domain generalization (SDG) aims to train a robust model against unknown target domain shifts using data from a single source domain. Data augmentation has been proven an effective approach to SDG. However, the utility of standard augmentations, such as translate, or invert, has not been fully exploited in SDG; practically, these augmentations are used as a part of a data preprocessing procedure. Although it is intuitive to use many such augmentations to boost the robustness of a model to out-of-distribution domain shifts, we lack a principled approach to harvest the benefit brought from multiple these augmentations. Here, we conceptualize standard data augmentations with learnable parameters as semantics transformations that can manipulate certain semantics of a sample, such as the geometry or color of an image. Then, we propose Adversarial learning with Semantics Transformations (AdvST) that augments the source domain data with semantics transformations and learns a robust model with the augmented data. We theoretically show that AdvST essentially optimizes a distributionally robust optimization objective defined on a set of semantics distributions induced by the parameters of semantics transformations. We demonstrate that AdvST can produce samples that expand the coverage on target domain data. Compared with the state-of-the-art methods, AdvST, despite being a simple method, is surprisingly competitive and achieves the best average SDG performance on the Digits, PACS, and DomainNet datasets. Our code is available at https://github.com/gtzheng/AdvST.

Sanchit Sinha, Guangzhi Xiong, Aidong Zhang

Interpretability of Deep Neural Networks using concept-based models offers a promising way to explain model behavior through human-understandable concepts. A parallel line of research focuses on disentangling the data distribution into its underlying generative factors, in turn explaining the data generation process. While both directions have received extensive attention, little work has been done on explaining concepts in terms of generative factors to unify mathematically disentangled representations and human-understandable concepts as an explanation for downstream tasks. In this paper, we propose a novel method CoLiDR - which utilizes a disentangled representation learning setup for learning mutually independent generative factors and subsequently learns to aggregate the said representations into human-understandable concepts using a novel aggregation/decomposition module. Experiments are conducted on datasets with both known and unknown latent generative factors. Our method successfully aggregates disentangled generative factors into concepts while maintaining parity with state-of-the-art concept-based approaches. Quantitative and visual analysis of the learned aggregation procedure demonstrates the advantages of our work compared to commonly used concept-based models over four challenging datasets. Lastly, our work is generalizable to an arbitrary number of concepts and generative factors - making it flexible enough to be suitable for various types of data.

Jianhui Sun, Sanchit Sinha, Aidong Zhang

Deep neural networks are susceptible to human imperceptible adversarial perturbations. One of the strongest defense mechanisms is \emph{Adversarial Training} (AT). In this paper, we aim to address two predominant problems in AT. First, there is still little consensus on how to set hyperparameters with a performance guarantee for AT research, and customized settings impede a fair comparison between different model designs in AT research. Second, the robustly trained neural networks struggle to generalize well and suffer from tremendous overfitting. This paper focuses on the primary AT framework - Projected Gradient Descent Adversarial Training (PGD-AT). We approximate the dynamic of PGD-AT by a continuous-time Stochastic Differential Equation (SDE), and show that the diffusion term of this SDE determines the robust generalization. An immediate implication of this theoretical finding is that robust generalization is positively correlated with the ratio between learning rate and batch size. We further propose a novel approach, \emph{Diffusion Enhanced Adversarial Training} (DEAT), to manipulate the diffusion term to improve robust generalization with virtually no extra computational burden. We theoretically show that DEAT obtains a tighter generalization bound than PGD-AT. Our empirical investigation is extensive and firmly attests that DEAT universally outperforms PGD-AT by a significant margin.

Peipei Zhou, Zheng Dong, Insup Lee et al.

This report summarizes the discussions and recommendations from the NSF Workshop on Algorithm-Hardware Co-design for Medical Applications, held on September 26-27, 2024, in Pittsburgh, PA. The workshop assembled an interdisciplinary cohort of researchers, clinicians, and industry leaders to examine foundational challenges and develop a strategic roadmap for algorithm-hardware co-design in medical computing. The workshop focuses on four thematic areas: (1) teleoperations, telehealth, and surgical operations; (2) wearable and implantable medicine, including implantable living pharmacies; (3) home ICU, hospital systems, and elderly care; and (4) medical sensing, imaging, and reconstruction. This report calls for a fundamental shift in how next-generation medical technologies are conceived, designed, validated, and translated into practice. The report recommends that NSF sustain investment in shared standardized data infrastructures and compute infrastructures, develop clinic workflow-aware systems and human-AI collaboration frameworks, promote scalable validation ecosystems grounded in objective, continuous measures, and physics-informed, and enable safe, accountable, and resilient platforms, including virtual-physical healthcare ecosystems, to de-risk translational pathways. The workshop information can be found on the website: https://sites.google.com/view/nsfworkshop.

Yin Fang, Qiao Jin, Guangzhi Xiong et al.

Cell type annotation is a key task in analyzing the heterogeneity of single-cell RNA sequencing data. Although recent foundation models automate this process, they typically annotate cells independently, without considering batch-level cellular context or providing explanatory reasoning. In contrast, human experts often annotate distinct cell types for different cell clusters based on their domain knowledge. To mimic this workflow, we introduce the CellPuzzles task, where the objective is to assign unique cell types to a batch of cells. This benchmark spans diverse tissues, diseases, and donor conditions, and requires reasoning across the batch-level cellular context to ensure label uniqueness. We find that off-the-shelf large language models (LLMs) struggle on CellPuzzles, with the best baseline (OpenAI's o1) achieving only 19.0% batch-level accuracy. To fill this gap, we propose Cell-o1, a 7B LLM trained via supervised fine-tuning on distilled reasoning traces, followed by reinforcement learning with batch-level rewards. Cell-o1 achieves state-of-the-art performance, outperforming o1 by over 73% and generalizing well across contexts. Further analysis of training dynamics and reasoning behaviors provides insights into batch-level annotation performance and emergent expert-like reasoning. Code and data are available at https://github.com/ncbi-nlp/cell-o1.

Guangzhi Xiong, Eric Xie, Corey Williams et al.

Large language models (LLMs) have shown significant potential in scientific disciplines such as biomedicine, particularly in hypothesis generation, where they can analyze vast literature, identify patterns, and suggest research directions. However, a key challenge lies in evaluating the truthfulness of generated hypotheses, as verifying their accuracy often requires substantial time and resources. Additionally, the hallucination problem in LLMs can lead to the generation of hypotheses that appear plausible but are ultimately incorrect, undermining their reliability. To facilitate the systematic study of these challenges, we introduce TruthHypo, a benchmark for assessing the capabilities of LLMs in generating truthful scientific hypotheses, and KnowHD, a knowledge-based hallucination detector to evaluate how well hypotheses are grounded in existing knowledge. Our results show that LLMs struggle to generate truthful hypotheses. By analyzing hallucinations in reasoning steps, we demonstrate that the groundedness scores provided by KnowHD serve as an effective metric for filtering truthful hypotheses from the diverse outputs of LLMs. Human evaluations further validate the utility of KnowHD in identifying truthful hypotheses and accelerating scientific discovery. Our data and source code are available at https://github.com/Teddy-XiongGZ/TruthHypo.

Sanchit Sinha, Guangzhi Xiong, Zhenghao He et al.

Modern vision-language models (VLMs) deliver impressive predictive accuracy yet offer little insight into 'why' a decision is reached, frequently hallucinating facts, particularly when encountering out-of-distribution data. Neurosymbolic frameworks address this by pairing black-box perception with interpretable symbolic reasoning, but current methods extract their symbols solely from task labels, leaving them weakly grounded in the underlying visual data. In this paper, we introduce a multi-agent system - Concept-RuleNet that reinstates visual grounding while retaining transparent reasoning. Specifically, a multimodal concept generator first mines discriminative visual concepts directly from a representative subset of training images. Next, these visual concepts are utilized to condition symbol discovery, anchoring the generations in real image statistics and mitigating label bias. Subsequently, symbols are composed into executable first-order rules by a large language model reasoner agent - yielding interpretable neurosymbolic rules. Finally, during inference, a vision verifier agent quantifies the degree of presence of each symbol and triggers rule execution in tandem with outputs of black-box neural models, predictions with explicit reasoning pathways. Experiments on five benchmarks, including two challenging medical-imaging tasks and three underrepresented natural-image datasets, show that our system augments state-of-the-art neurosymbolic baselines by an average of 5% while also reducing the occurrence of hallucinated symbols in rules by up to 50%.

Guangzhi Xiong, Qiao Jin, Zhiyong Lu et al.

While large language models (LLMs) have achieved state-of-the-art performance on a wide range of medical question answering (QA) tasks, they still face challenges with hallucinations and outdated knowledge. Retrieval-augmented generation (RAG) is a promising solution and has been widely adopted. However, a RAG system can involve multiple flexible components, and there is a lack of best practices regarding the optimal RAG setting for various medical purposes. To systematically evaluate such systems, we propose the Medical Information Retrieval-Augmented Generation Evaluation (MIRAGE), a first-of-its-kind benchmark including 7,663 questions from five medical QA datasets. Using MIRAGE, we conducted large-scale experiments with over 1.8 trillion prompt tokens on 41 combinations of different corpora, retrievers, and backbone LLMs through the MedRAG toolkit introduced in this work. Overall, MedRAG improves the accuracy of six different LLMs by up to 18% over chain-of-thought prompting, elevating the performance of GPT-3.5 and Mixtral to GPT-4-level. Our results show that the combination of various medical corpora and retrievers achieves the best performance. In addition, we discovered a log-linear scaling property and the "lost-in-the-middle" effects in medical RAG. We believe our comprehensive evaluations can serve as practical guidelines for implementing RAG systems for medicine.

Qiao Jin, Yin Fang, Lauren He et al.

Assessing whether an article supports an assertion is essential for hallucination detection and claim verification. While large language models (LLMs) have the potential to automate this task, achieving strong performance requires frontier models such as GPT-5 that are prohibitively expensive to deploy at scale. To efficiently perform biomedical evidence attribution, we present Med-V1, a family of small language models with only three billion parameters. Trained on high-quality synthetic data newly developed in this study, Med-V1 substantially outperforms (+27.0% to +71.3%) its base models on five biomedical benchmarks unified into a verification format. Despite its smaller size, Med-V1 performs comparably to frontier LLMs such as GPT-5, along with high-quality explanations for its predictions. We use Med-V1 to conduct a first-of-its-kind use case study that quantifies hallucinations in LLM-generated answers under different citation instructions. Results show that the format instruction strongly affects citation validity and hallucination, with GPT-5 generating more claims but exhibiting hallucination rates similar to GPT-4o. Additionally, we present a second use case showing that Med-V1 can automatically identify high-stakes evidence misattributions in clinical practice guidelines, revealing potentially negative public health impacts that are otherwise challenging to identify at scale. Overall, Med-V1 provides an efficient and accurate lightweight alternative to frontier LLMs for practical and real-world applications in biomedical evidence attribution and verification tasks. Med-V1 is available at https://github.com/ncbi-nlp/Med-V1.

Xinyu Liu, Zixuan Xie, Amir Moeini et al.

While the ecosystem of Lean and Mathlib has enjoyed celebrated success in formal mathematical reasoning with the help of large language models (LLMs), the absence of many folklore lemmas in Mathlib remains a persistent barrier that limits Lean's usability as an everyday tool for mathematicians like LaTeX or Maple. To address this, we introduce MathlibLemma, the first LLM-based multi-agent system to automate the discovery and formalization of mathematical folklore lemmas. This framework constitutes our primary contribution, proactively mining the missing connective tissue of mathematics. Its efficacy is demonstrated by the production of a verified library of folklore lemmas, a subset of which has already been formally merged into the latest build of Mathlib, thereby validating the system's real-world utility and alignment with expert standards. Leveraging this pipeline, we further construct the MathlibLemma benchmark, a suite of 4,028 type-checked Lean statements spanning a broad range of mathematical domains. By transforming the role of LLMs from passive consumers to active contributors, this work establishes a constructive methodology for the self-evolution of formal mathematical libraries.

Sanchit Sinha, Guangzhi Xiong, Aidong Zhang

Compositional reasoning remains a persistent weakness of modern vision language models (VLMs): they often falter when a task hinges on understanding how multiple objects, attributes, and relations interact within an image. Multiple research works have attempted to improve compositionality performance by creative tricks such as improving prompt structure, chain of thought reasoning, etc. A more recent line of work attempts to impart additional reasoning in VLMs using well-trained Large Language Models (LLMs), which are far superior in linguistic understanding than VLMs to compensate for the limited linguistic prowess of VLMs. However, these approaches are either resource-intensive or do not provide an interpretable reasoning process. In this paper, we present 'COCO-Tree' - a novel approach that augments VLM outputs with carefully designed neurosymbolic concept trees learned from LLMs to improve VLM's linguistic reasoning. COCO-Tree's beam search-inspired reasoning process boosts compositionality performance and provides a rationale behind VLM predictions. Empirical results on four compositionality benchmarks, Winoground, EqBench, ColorSwap, and SugarCrepe, in seven different open-source VLMs with varying sizes, demonstrate that COCO-Tree significantly improves compositional generalization by 5-10% over baselines.

Zhenghao He, Sanchit Sinha, Guangzhi Xiong et al.

Concept Activation Vectors (CAVs) provide a powerful approach for interpreting deep neural networks by quantifying their sensitivity to human-defined concepts. However, when computed independently at different layers, CAVs often exhibit inconsistencies, making cross-layer comparisons unreliable. To address this issue, we propose the Global Concept Activation Vector (GCAV), a novel framework that unifies CAVs into a single, semantically consistent representation. Our method leverages contrastive learning to align concept representations across layers and employs an attention-based fusion mechanism to construct a globally integrated CAV. By doing so, our method significantly reduces the variance in TCAV scores while preserving concept relevance, ensuring more stable and reliable concept attributions. To evaluate the effectiveness of GCAV, we introduce Testing with Global Concept Activation Vectors (TGCAV) as a method to apply TCAV to GCAV-based representations. We conduct extensive experiments on multiple deep neural networks, demonstrating that our method effectively mitigates concept inconsistency across layers, enhances concept localization, and improves robustness against adversarial perturbations. By integrating cross-layer information into a coherent framework, our method offers a more comprehensive and interpretable understanding of how deep learning models encode human-defined concepts. Code and models are available at https://github.com/Zhenghao-He/GCAV.

Wenqian Ye, Bohan Liu, Guangtao Zheng et al.

Spurious bias, a tendency to exploit spurious correlations between superficial input attributes and prediction targets, has revealed a severe robustness pitfall in classical machine learning problems. Multimodal Large Language Models (MLLMs), which leverage pretrained vision and language models, have recently demonstrated strong capability in joint vision-language understanding. However, both the presence and severity of spurious biases in MLLMs remain poorly understood. In this work, we address this gap by analyzing the spurious biases in the multimodal setting and uncovering the specific inference-time data patterns that can manifest this problem. To support this analysis, we introduce MM-SpuBench, a comprehensive, human-verified benchmark dataset consisting of image-class pairs annotated with core and spurious attributes, grounded in our taxonomy of nine distinct types of spurious correlations. The benchmark is constructed using human-interpretable attribute information to capture a wide range of spurious patterns reflective of real-world knowledge. Leveraging this benchmark, we conduct a comprehensive evaluation of the state-of-the-art open-source and proprietary MLLMs with both standard accuracy and the proposed Conditional Generation Likelihood Advantage (CGLA). Our findings highlight the persistence of reliance on spurious correlations and the difficulty of mitigation on our benchmark. We hope this work can inspire new technical strides to mitigate these biases. Our benchmark is publicly available at https://huggingface.co/datasets/mmbench/MM-SpuBench.

Xu Cao, Yifan Shen, Bolin Lai et al.

Recently, Multimodal Large Language Models (MLLMs) and Vision Language Models (VLMs) have shown great promise in language-guided perceptual tasks such as recognition, segmentation, and object detection. However, their effectiveness in addressing visual cognition problems that require high-level multi-image reasoning and visual working memory is not well-established. One such challenge is matrix reasoning - the cognitive ability to discern relationships among patterns in a set of images and extrapolate to predict subsequent patterns. This skill is crucial during the early neurodevelopmental stages of children. Inspired by the matrix reasoning tasks in Raven's Progressive Matrices (RPM) and Wechsler Intelligence Scale for Children (WISC), we propose a new dataset MaRs-VQA to evaluate the visual cognition capability of MLLMs and compare their performance with existing human visual cognition studies. Based on the training data of MaRs-VQA, we also finetune a baseline model Qwen2-VCog with multi-stage cognition reasoning annotations. Our comparative experiments with different baselines reveal a gap between MLLMs and human intelligence, highlighting the visual cognitive limitations of current MLLMs. We believe that the public release of MaRs-VQA and the Qwen2-VCog baseline model will drive progress toward the next generation of MLLMs with human-like visual cognition abilities. MaRs-VQA is available at huggingface.co/datasets/IrohXu/VCog-Bench. The training code of Qwen2-VCog is available at github.com/IrohXu/Cognition-MLLM.

Nicholas Wan, Qiao Jin, Joey Chan et al.

Although large language models (LLMs) have been assessed for general medical knowledge using licensing exams, their ability to support clinical decision-making, such as selecting medical calculators, remains uncertain. We assessed nine LLMs, including open-source, proprietary, and domain-specific models, with 1,009 multiple-choice question-answer pairs across 35 clinical calculators and compared LLMs to humans on a subset of questions. While the highest-performing LLM, OpenAI o1, provided an answer accuracy of 66.0% (CI: 56.7-75.3%) on the subset of 100 questions, two human annotators nominally outperformed LLMs with an average answer accuracy of 79.5% (CI: 73.5-85.0%). Ultimately, we evaluated medical trainees and LLMs in recommending medical calculators across clinical scenarios like risk stratification and diagnosis. With error analysis showing that the highest-performing LLMs continue to make mistakes in comprehension (49.3% of errors) and calculator knowledge (7.1% of errors), our findings highlight that LLMs are not superior to humans in calculator recommendation.

Liuyi Yao, Zhixuan Chu, Sheng Li et al.

Causal inference is a critical research topic across many domains, such as statistics, computer science, education, public policy and economics, for decades. Nowadays, estimating causal effect from observational data has become an appealing research direction owing to the large amount of available data and low budget requirement, compared with randomized controlled trials. Embraced with the rapidly developed machine learning area, various causal effect estimation methods for observational data have sprung up. In this survey, we provide a comprehensive review of causal inference methods under the potential outcome framework, one of the well known causal inference framework. The methods are divided into two categories depending on whether they require all three assumptions of the potential outcome framework or not. For each category, both the traditional statistical methods and the recent machine learning enhanced methods are discussed and compared. The plausible applications of these methods are also presented, including the applications in advertising, recommendation, medicine and so on. Moreover, the commonly used benchmark datasets as well as the open-source codes are also summarized, which facilitate researchers and practitioners to explore, evaluate and apply the causal inference methods.

Tianle Ma, Aidong Zhang

While deep learning has achieved great success in computer vision and many other fields, currently it does not work very well on patient genomic data with the "big p, small N" problem (i.e., a relatively small number of samples with high-dimensional features). In order to make deep learning work with a small amount of training data, we have to design new models that facilitate few-shot learning. Here we present the Affinity Network Model (AffinityNet), a data efficient deep learning model that can learn from a limited number of training examples and generalize well. The backbone of the AffinityNet model consists of stacked k-Nearest-Neighbor (kNN) attention pooling layers. The kNN attention pooling layer is a generalization of the Graph Attention Model (GAM), and can be applied to not only graphs but also any set of objects regardless of whether a graph is given or not. As a new deep learning module, kNN attention pooling layers can be plugged into any neural network model just like convolutional layers. As a simple special case of kNN attention pooling layer, feature attention layer can directly select important features that are useful for classification tasks. Experiments on both synthetic data and cancer genomic data from TCGA projects show that our AffinityNet model has better generalization power than conventional neural network models with little training data. The code is freely available at https://github.com/BeautyOfWeb/AffinityNet .

Wenqian Ye, Luyang Jiang, Eric Xie et al.

Back in the early 20th century, a horse named Hans appeared to perform arithmetic and other intellectual tasks during exhibitions in Germany, while it actually relied solely on involuntary cues in the body language from the human trainer. Modern machine learning models are no different. These models are known to be sensitive to spurious correlations between non-essential features of the inputs (e.g., background, texture, and secondary objects) and the corresponding labels. Such features and their correlations with the labels are known as "spurious" because they tend to change with shifts in real-world data distributions, which can negatively impact the model's generalization and robustness. In this paper, we provide a comprehensive survey of this emerging issue, along with a fine-grained taxonomy of existing state-of-the-art methods for addressing spurious correlations in machine learning models. Additionally, we summarize existing datasets, benchmarks, and metrics to facilitate future research. The paper concludes with a discussion of the broader impacts, the recent advancements, and future challenges in the era of generative AI, aiming to provide valuable insights for researchers in the related domains of the machine learning community.

Eric Xie, Danielle Waterfield, Michael Kennedy et al.

Large Language Models (LLMs) have shown immense potential in education, automating tasks like quiz generation and content summarization. However, generating effective presentation slides introduces unique challenges due to the complexity of multimodal content creation and the need for precise, domain-specific information. Existing LLM-based solutions often fail to produce reliable and informative outputs, limiting their educational value. To address these limitations, we introduce SlideBot - a modular, multi-agent slide generation framework that integrates LLMs with retrieval, structured planning, and code generation. SlideBot is organized around three pillars: informativeness, ensuring deep and contextually grounded content; reliability, achieved by incorporating external sources through retrieval; and practicality, which enables customization and iterative feedback through instructor collaboration. It incorporates evidence-based instructional design principles from Cognitive Load Theory (CLT) and the Cognitive Theory of Multimedia Learning (CTML), using structured planning to manage intrinsic load and consistent visual macros to reduce extraneous load and enhance dual-channel learning. Within the system, specialized agents collaboratively retrieve information, summarize content, generate figures, and format slides using LaTeX, aligning outputs with instructor preferences through interactive refinement. Evaluations from domain experts and students in AI and biomedical education show that SlideBot consistently enhances conceptual accuracy, clarity, and instructional value. These findings demonstrate SlideBot's potential to streamline slide preparation while ensuring accuracy, relevance, and adaptability in higher education.

Sikun Guo, Amir Hassan Shariatmadari, Guangzhi Xiong et al.

Large Language Models (LLMs) have transformed how people interact with artificial intelligence (AI) systems, achieving state-of-the-art results in various tasks, including scientific discovery and hypothesis generation. However, the lack of a comprehensive and systematic evaluation framework for generating research ideas using LLMs poses a significant obstacle to understanding and assessing their generative capabilities in scientific discovery. To address this gap, we propose IdeaBench, a benchmark system that includes a comprehensive dataset and an evaluation framework for standardizing the assessment of research idea generation using LLMs. Our dataset comprises titles and abstracts from a diverse range of influential papers, along with their referenced works. To emulate the human process of generating research ideas, we profile LLMs as domain-specific researchers and ground them in the same context considered by human researchers. This maximizes the utilization of the LLMs' parametric knowledge to dynamically generate new research ideas. We also introduce an evaluation framework for assessing the quality of generated research ideas. Our evaluation framework is a two-stage process: first, using GPT-4o to rank ideas based on user-specified quality indicators such as novelty and feasibility, enabling scalable personalization; and second, calculating relative ranking based "Insight Score" to quantify the chosen quality indicator. The proposed benchmark system will be a valuable asset for the community to measure and compare different LLMs, ultimately advancing the automation of the scientific discovery process.

Guangzhi Xiong, Eric Xie, Amir Hassan Shariatmadari et al.

Large language models (LLMs) have demonstrated remarkable capabilities in various scientific domains, from natural language processing to complex problem-solving tasks. Their ability to understand and generate human-like text has opened up new possibilities for advancing scientific research, enabling tasks such as data analysis, literature review, and even experimental design. One of the most promising applications of LLMs in this context is hypothesis generation, where they can identify novel research directions by analyzing existing knowledge. However, despite their potential, LLMs are prone to generating ``hallucinations'', outputs that are plausible-sounding but factually incorrect. Such a problem presents significant challenges in scientific fields that demand rigorous accuracy and verifiability, potentially leading to erroneous or misleading conclusions. To overcome these challenges, we propose KG-CoI (Knowledge Grounded Chain of Ideas), a novel system that enhances LLM hypothesis generation by integrating external, structured knowledge from knowledge graphs (KGs). KG-CoI guides LLMs through a structured reasoning process, organizing their output as a chain of ideas (CoI), and includes a KG-supported module for the detection of hallucinations. With experiments on our newly constructed hypothesis generation dataset, we demonstrate that KG-CoI not only improves the accuracy of LLM-generated hypotheses but also reduces the hallucination in their reasoning chains, highlighting its effectiveness in advancing real-world scientific research.

Jianhui Sun, Xidong Wu, Heng Huang et al.

Federated Averaging (FedAvg) is known to experience convergence issues when encountering significant clients system heterogeneity and data heterogeneity. Server momentum has been proposed as an effective mitigation. However, existing server momentum works are restrictive in the momentum formulation, do not properly schedule hyperparameters and focus only on system homogeneous settings, which leaves the role of server momentum still an under-explored problem. In this paper, we propose a general framework for server momentum, that (a) covers a large class of momentum schemes that are unexplored in federated learning (FL), (b) enables a popular stagewise hyperparameter scheduler, (c) allows heterogeneous and asynchronous local computing. We provide rigorous convergence analysis for the proposed framework. To our best knowledge, this is the first work that thoroughly analyzes the performances of server momentum with a hyperparameter scheduler and system heterogeneity. Extensive experiments validate the effectiveness of our proposed framework.

Sanchit Sinha, Yuguang Yue, Victor Soto et al.

Adapting large language models (LLMs) to unseen tasks with in-context training samples without fine-tuning remains an important research problem. To learn a robust LLM that adapts well to unseen tasks, multiple meta-training approaches have been proposed such as MetaICL and MetaICT, which involve meta-training pre-trained LLMs on a wide variety of diverse tasks. These meta-training approaches essentially perform in-context multi-task fine-tuning and evaluate on a disjointed test set of tasks. Even though they achieve impressive performance, their goal is never to compute a truly general set of parameters. In this paper, we propose MAML-en-LLM, a novel method for meta-training LLMs, which can learn truly generalizable parameters that not only perform well on disjointed tasks but also adapts to unseen tasks. We see an average increase of 2% on unseen domains in the performance while a massive 4% improvement on adaptation performance. Furthermore, we demonstrate that MAML-en-LLM outperforms baselines in settings with limited amount of training data on both seen and unseen domains by an average of 2%. Finally, we discuss the effects of type of tasks, optimizers and task complexity, an avenue barely explored in meta-training literature. Exhaustive experiments across 7 task settings along with two data settings demonstrate that models trained with MAML-en-LLM outperform SOTA meta-training approaches.

Qiao Jin, Nicholas Wan, Robert Leaman et al.

Large language models (LLMs) represent a transformative class of AI tools capable of revolutionizing various aspects of healthcare by generating human-like responses across diverse contexts and adapting to novel tasks following human instructions. Their potential application spans a broad range of medical tasks, such as clinical documentation, matching patients to clinical trials, and answering medical questions. In this primer paper, we propose an actionable guideline to help healthcare professionals more efficiently utilize LLMs in their work, along with a set of best practices. This approach consists of several main phases, including formulating the task, choosing LLMs, prompt engineering, fine-tuning, and deployment. We start with the discussion of critical considerations in identifying healthcare tasks that align with the core capabilities of LLMs and selecting models based on the selected task and data, performance requirements, and model interface. We then review the strategies, such as prompt engineering and fine-tuning, to adapt standard LLMs to specialized medical tasks. Deployment considerations, including regulatory compliance, ethical guidelines, and continuous monitoring for fairness and bias, are also discussed. By providing a structured step-by-step methodology, this tutorial aims to equip healthcare professionals with the tools necessary to effectively integrate LLMs into clinical practice, ensuring that these powerful technologies are applied in a safe, reliable, and impactful manner.

Guangzhi Xiong, Qiao Jin, Xiao Wang et al.

Retrieval-augmented generation (RAG) has shown great promise for knowledge-intensive tasks and recently advanced with agentic RAG, where language agents engage in multi-round interactions with external knowledge sources for adaptive information retrieval. However, existing agentic RAG methods often depend on ad-hoc prompt engineering and lack a unified optimization framework. We introduce RAG-Gym, a comprehensive platform that systematically explores three optimization dimensions: (1) prompt engineering, (2) actor tuning, and (3) critic training. For prompt engineering, we propose Re$^2$Search, a novel agent incorporating reasoning reflection that significantly outperforms standard prompts. In actor tuning, we evaluate three popular post-training algorithms with fine-grained process supervision and identify direct preference optimization as the most effective. We further demonstrate that a trained critic can enhance inference by selecting higher-quality intermediate reasoning steps. Together, these findings lead to the optimized Re$^2$Search++ agent, which surpasses most recent methods like Search-R1 by a relative increase of 3.2% to 11.6% in average F1. Finally, we examine the impact of different reward sources and analyze scaling properties in training and inference, offering practical insights for agentic RAG optimization. The project homepage is available at https://rag-gym.github.io.

Yifan Yang, Qiao Jin, Robert Leaman et al.

The remarkable capabilities of Large Language Models (LLMs) make them increasingly compelling for adoption in real-world healthcare applications. However, the risks associated with using LLMs in medical applications have not been systematically characterized. We propose using five key principles for safe and trustworthy medical AI: Truthfulness, Resilience, Fairness, Robustness, and Privacy, along with ten specific aspects. Under this comprehensive framework, we introduce a novel MedGuard benchmark with 1,000 expert-verified questions. Our evaluation of 11 commonly used LLMs shows that the current language models, regardless of their safety alignment mechanisms, generally perform poorly on most of our benchmarks, particularly when compared to the high performance of human physicians. Despite recent reports indicate that advanced LLMs like ChatGPT can match or even exceed human performance in various medical tasks, this study underscores a significant safety gap, highlighting the crucial need for human oversight and the implementation of AI safety guardrails.

Guangtao Zheng, Wenqian Ye, Aidong Zhang

Deep learning models often achieve high performance by inadvertently learning spurious correlations between targets and non-essential features. For example, an image classifier may identify an object via its background that spuriously correlates with it. This prediction behavior, known as spurious bias, severely degrades model performance on data that lacks the learned spurious correlations. Existing methods on spurious bias mitigation typically require a variety of data groups with spurious correlation annotations called group labels. However, group labels require costly human annotations and often fail to capture subtle spurious biases such as relying on specific pixels for predictions. In this paper, we propose a novel post hoc spurious bias mitigation framework without requiring group labels. Our framework, termed ShortcutProbe, identifies prediction shortcuts that reflect potential non-robustness in predictions in a given model's latent space. The model is then retrained to be invariant to the identified prediction shortcuts for improved robustness. We theoretically analyze the effectiveness of the framework and empirically demonstrate that it is an efficient and practical tool for improving a model's robustness to spurious bias on diverse datasets.

Sanchit Sinha, Guangzhi Xiong, Aidong Zhang

With the wide proliferation of Deep Neural Networks in high-stake applications, there is a growing demand for explainability behind their decision-making process. Concept learning models attempt to learn high-level 'concepts' - abstract entities that align with human understanding, and thus provide interpretability to DNN architectures. However, in this paper, we demonstrate that present SOTA concept learning approaches suffer from two major problems - lack of concept fidelity wherein the models fail to learn consistent concepts among similar classes and limited concept interoperability wherein the models fail to generalize learned concepts to new domains for the same task. Keeping these in mind, we propose a novel self-explaining architecture for concept learning across domains which - i) incorporates a new concept saliency network for representative concept selection, ii) utilizes contrastive learning to capture representative domain invariant concepts, and iii) uses a novel prototype-based concept grounding regularization to improve concept alignment across domains. We demonstrate the efficacy of our proposed approach over current SOTA concept learning approaches on four widely used real-world datasets. Empirical results show that our method improves both concept fidelity measured through concept overlap and concept interoperability measured through domain adaptation performance.

Wenqian Ye, Guangtao Zheng, Aidong Zhang

Deep neural networks often learn and rely on spurious correlations, i.e., superficial associations between non-causal features and the targets. For instance, an image classifier may identify camels based on the desert backgrounds. While it can yield high overall accuracy during training, it degrades generalization on more diverse scenarios where such correlations do not hold. This problem poses significant challenges for out-of-distribution robustness and trustworthiness. Existing methods typically mitigate this issue by using external group annotations or auxiliary deterministic models to learn unbiased representations. However, such information is costly to obtain, and deterministic models may fail to capture the full spectrum of biases learned by the models. To address these limitations, we propose Evidential Alignment, a novel framework that leverages uncertainty quantification to understand the behavior of the biased models without requiring group annotations. By quantifying the evidence of model prediction with second-order risk minimization and calibrating the biased models with the proposed evidential calibration technique, Evidential Alignment identifies and suppresses spurious correlations while preserving core features. We theoretically justify the effectiveness of our method as capable of learning the patterns of biased models and debiasing the model without requiring any spurious correlation annotations. Empirical results demonstrate that our method significantly improves group robustness across diverse architectures and data modalities, providing a scalable and principled solution to spurious correlations.

Guangtao Zheng, Wenqian Ye, Aidong Zhang

Deep neural networks often develop spurious bias, reliance on correlations between non-essential features and classes for predictions. For example, a model may identify objects based on frequently co-occurring backgrounds rather than intrinsic features, resulting in degraded performance on data lacking these correlations. Existing mitigation approaches typically depend on external annotations of spurious correlations, which may be difficult to obtain and are not relevant to the spurious bias in a model. In this paper, we take a step towards self-guided mitigation of spurious bias by proposing NeuronTune, a post hoc method that directly intervenes in a model's internal decision process. Our method probes in a model's latent embedding space to identify and regulate neurons that lead to spurious prediction behaviors. We theoretically justify our approach and show that it brings the model closer to an unbiased one. Unlike previous methods, NeuronTune operates without requiring spurious correlation annotations, making it a practical and effective tool for improving model robustness. Experiments across different architectures and data modalities demonstrate that our method significantly mitigates spurious bias in a self-guided way.

Huimin Xu, Seungjun Yi, Terence Lim et al.

Thematic analysis (TA) is a widely used qualitative approach for uncovering latent meanings in unstructured text data. TA provides valuable insights in healthcare but is resource-intensive. Large Language Models (LLMs) have been introduced to perform TA, yet their applications in healthcare remain unexplored. Here, we propose TAMA: A Human-AI Collaborative Thematic Analysis framework using Multi-Agent LLMs for clinical interviews. We leverage the scalability and coherence of multi-agent systems through structured conversations between agents and coordinate the expertise of cardiac experts in TA. Using interview transcripts from parents of children with Anomalous Aortic Origin of a Coronary Artery (AAOCA), a rare congenital heart disease, we demonstrate that TAMA outperforms existing LLM-assisted TA approaches, achieving higher thematic hit rate, coverage, and distinctiveness. TAMA demonstrates strong potential for automated TA in clinical settings by leveraging multi-agent LLM systems with human-in-the-loop integration by enhancing quality while significantly reducing manual workload.

Sanchit Sinha, Guangzhi Xiong, Aidong Zhang

As Vision Transformers (ViTs) are increasingly adopted in sensitive vision applications, there is a growing demand for improved interpretability. This has led to efforts to forward-align these models with carefully annotated abstract, human-understandable semantic entities - concepts. Concepts provide global rationales to the model predictions and can be quickly understood/intervened on by domain experts. Most current research focuses on designing model-agnostic, plug-and-play generic concept-based explainability modules that do not incorporate the inner workings of foundation models (e.g., inductive biases, scale invariance, etc.) during training. To alleviate this issue for ViTs, in this paper, we propose ASCENT-ViT, an attention-based, concept learning framework that effectively composes scale and position-aware representations from multiscale feature pyramids and ViT patch representations, respectively. Further, these representations are aligned with concept annotations through attention matrices - which incorporate spatial and global (semantic) concepts. ASCENT-ViT can be utilized as a classification head on top of standard ViT backbones for improved predictive performance and accurate and robust concept explanations as demonstrated on five datasets, including three widely used benchmarks (CUB, Pascal APY, Concept-MNIST) and 2 real-world datasets (AWA2, KITS).

Zhenghao He, Guangzhi Xiong, Boyang Wang et al.

Internal activations of diffusion models encode rich semantic information, but interpreting such representations remains challenging. While Sparse Autoencoders (SAEs) have shown promise in disentangling latent representations, existing SAE-based methods for diffusion model understanding rely on unsupervised approaches that fail to align sparse features with human-understandable concepts. This limits their ability to provide reliable semantic control over generated images. We introduce CASL (Concept-Aligned Sparse Latents), a supervised framework that aligns sparse latent dimensions of diffusion models with semantic concepts. CASL first trains an SAE on frozen U-Net activations to obtain disentangled latent representations, and then learns a lightweight linear mapping that associates each concept with a small set of relevant latent dimensions. To validate the semantic meaning of these aligned directions, we propose CASL-Steer, a controlled latent intervention that shifts activations along the learned concept axis. Unlike editing methods, CASL-Steer is used solely as a causal probe to reveal how concept-aligned latents influence generated content. We further introduce the Editing Precision Ratio (EPR), a metric that jointly measures concept specificity and the preservation of unrelated attributes. Experiments show that our method achieves superior editing precision and interpretability compared to existing approaches. To the best of our knowledge, this is the first work to achieve supervised alignment between latent representations and semantic concepts in diffusion models.

Zhenghao He, Guangzhi Xiong, Bohan Liu et al.

Chain-of-Thought (CoT) prompting has improved the reasoning performance of large language models (LLMs), but it remains unclear why it works and whether it is the unique mechanism for triggering reasoning in large language models. In this work, we study this question by directly analyzing and intervening on the internal representations of LLMs with Sparse Autoencoders (SAEs), identifying a small set of latent features that are causally associated with LLM reasoning behavior. Across multiple model families and reasoning benchmarks, we find that steering a single reasoning-related latent feature can substantially improve accuracy without explicit CoT prompting. For large models, latent steering achieves performance comparable to standard CoT prompting while producing more efficient outputs. We further observe that this reasoning-oriented internal state is triggered early in generation and can override prompt-level instructions that discourage explicit reasoning. Overall, our results suggest that multi-step reasoning in LLMs is supported by latent internal activations that can be externally activated, while CoT prompting is one effective, but not unique, way of activating this mechanism rather than its necessary cause.

Wenqian Ye, Di Wang, Guangtao Zheng et al.

Large vision-language models, such as CLIP, have shown strong zero-shot classification performance by aligning images and text in a shared embedding space. However, CLIP models often develop multimodal spurious biases, which is the undesirable tendency to rely on spurious features. For example, CLIP may infer object types in images based on frequently co-occurring backgrounds rather than the object's core features. This bias significantly impairs the robustness of pre-trained CLIP models on out-of-distribution data, where such cross-modal associations no longer hold. Existing methods for mitigating multimodal spurious bias typically require fine-tuning on downstream data or prior knowledge of the bias, which undermines the out-of-the-box usability of CLIP. In this paper, we first theoretically analyze the impact of multimodal spurious bias in zero-shot classification. Based on this insight, we propose Spuriousness-Aware Guided Exploration (SAGE), a simple and effective method that mitigates spurious bias through guided prompt selection. SAGE requires no training, fine-tuning, or external annotations. It explores a space of prompt templates and selects the prompts that induce the largest semantic separation between classes, thereby improving worst-group robustness. Extensive experiments on four real-world benchmark datasets and five popular backbone models demonstrate that SAGE consistently improves zero-shot performance and generalization, outperforming previous zero-shot approaches without any external knowledge or model updates.

Wenqian Ye, Guangtao Zheng, Aidong Zhang

In reinforcement learning from human feedback, preference-based reward models play a central role in aligning large language models to human-aligned behavior. However, recent studies show that these models are prone to reward hacking and often fail to generalize well due to over-optimization. They achieve high reward scores by exploiting shortcuts, that is, exploiting spurious features (e.g., response verbosity, agreeable tone, or sycophancy) that correlate with human preference labels in the training data rather than genuinely reflecting the intended objectives. In this paper, instead of probing these issues one at a time, we take a broader view of the reward hacking problem as shortcut behaviors and introduce a principled yet flexible approach to mitigate shortcut behaviors in preference-based reward learning. Inspired by the invariant theory in the kernel perspective, we propose Preference-based Reward Invariance for Shortcut Mitigation (PRISM), which learns group-invariant kernels with feature maps in a closed-form learning objective. Experimental results in several benchmarks show that our method consistently improves the accuracy of the reward model on diverse out-of-distribution tasks and reduces the dependency on shortcuts in downstream policy models, establishing a robust framework for preference-based alignment.