arXiv subjects

Saeed Hassanpour

Publications and source records attributed to Saeed Hassanpour.

At least 19 recordsLinked to original sources

arXiv2026

When Linguistic and Internal Confidence Diverge in Large Language Models

Users often ask large language models (LLMs) to report how confident they are, but it is unclear whether such linguistic confidence tracks the model's internal confidence. We study this question across 8 classification tasks, 2 generation tasks and 30 models from three families. For classification, we compare linguistic confidence with logits-based confidence along three axes: association, magnitude agreement and calibration. For generation, we test whether linguistic confidence tracks semantic-entropy-based uncertainty. The axes frequently diverge. Instance-level association is weak on average, although it improves on easier items and for stronger base models. Instruction-tuned models often report higher confidence and sometimes show higher association, but they also have larger confidence gaps and worse calibration. Prompt design mostly changes the distribution of reported confidence. Attitude cues inflate confidence without improving alignment, while score exemplars can preserve rank-order signal when they avoid collapsed confidence values. Regression analyses show that distributional properties of confidence scores explain much of the observed alignment pattern, with model metadata playing a smaller role after controls. These results support a lossy-channel view of linguistic confidence. A more dispersed verbal confidence distribution can carry useful rank information, but it does not make the scores calibrated. Linguistic confidence should therefore be evaluated with multi-axis diagnostics before being used in downstream reliability pipelines.

cs.CL↗

arXiv2026

Memory Makes the Difference: Evaluating How Different Memory Roles Shape Conversational Agents

Prior research on memory mechanism in RAG-based conversational system has emphasized how memory is stored and retrieved. However, far less is known about how memories with different functional roles influence response quality. Specifically, how they shape an agent's responses under varying conversational contexts and whether they lead to substantively different response behaviors. Existing evaluations in conversational system are also largely reference-based, insufficiently capturing the nuances in responses that may address users' preferences differently. In this work, we probe the impact of different memory types in shaping agents' responses. We present a fine-grained taxonomy of conversational memory, classify retrieved memories into different role types, and design a user-centric evaluation framework that simulates user perspectives. Through comparative experiments on long-term datasets and frontier LLMs, our analysis reveal many differentiated effects of memories: e.g., clarifying memory improves responses' factual accuracy and constraint awareness, making them more correct and personalized; irrelevant memory reduces topic relevance and degrades constraint awareness. Despite the power of frontier LLMs, these findings shed light on how different memory types can be leveraged to produce more personalized responses and inspire further research in this direction.

cs.CL↗

arXiv2026

Probing Association Biases in LLM Moderation Over-Sensitivity

Large Language Models are widely used for content moderation but often present certain over-sensitivity, leading to misclassification of benign content and rejecting safe user commands. While previous research attributes this issue primarily to the presence of explicit offensive triggers, we statistically reveal a deeper connection beyond token level: When behaving over-sensitively, particularly on decontextualized statements, LLMs exhibit systematic topic-toxicity association patterns that go beyond explicit offensive triggers. To characterize these patterns, we propose Topic Association Analysis, a behavior-based probe that elicits short contextual scenarios for benign inputs and quantifies topic amplification between the scenario and the original comment. Across multiple LLMs and large-scale data, we find that more advanced models (e.g., GPT-4 Turbo) show stronger topic-association skew in false-positive cases despite lower overall false-positive rates. Moreover, via controlled prefix interventions, we show that topic cues can measurably shift false-positive rates, indicating that topic framing is decision-relevant. These results suggest that mitigating over-sensitivity may require addressing learned topic associations in addition to keyword-based filtering.

cs.CL↗

arXiv2026

Cluster-R1: Large Reasoning Models Are Instruction-following Clustering Agents

General-purpose embedding models excel at recognizing semantic similarities but fail to capture the characteristics of texts specified by user instructions. In contrast, instruction-tuned embedders can align embeddings with textual instructions yet cannot autonomously infer latent corpus structures, such as determining the optimal number of clusters. To address both limitations, we reframe instruction-following clustering as a generative task and train large reasoning models (LRMs) as autonomous clustering agents. Our reasoning-driven training pipeline enables LRMs to interpret high-level clustering instructions and then infer the corresponding latent groupings. To evaluate this paradigm, we introduce ReasonCluster, a comprehensive benchmark comprising 28 diverse tasks spanning daily dialogue, legal cases, and financial reports. Experiments across diverse datasets and clustering scenarios show that our approach consistently outperforms strong embedding-based methods and LRM baselines, demonstrating that explicit reasoning fosters more faithful and interpretable instruction-based clustering.

cs.CL↗

arXiv2025

A Generalizable Rhetorical Strategy Annotation Model Using LLM-based Debate Simulation and Labelling

Rhetorical strategies are central to persuasive communication, from political discourse and marketing to legal argumentation. However, analysis of rhetorical strategies has been limited by reliance on human annotation, which is costly, inconsistent, difficult to scale. Their associated datasets are often limited to specific topics and strategies, posing challenges for robust model development. We propose a novel framework that leverages large language models (LLMs) to automatically generate and label synthetic debate data based on a four-part rhetorical typology (causal, empirical, emotional, moral). We fine-tune transformer-based classifiers on this LLM-labeled dataset and validate its performance against human-labeled data on this dataset and on multiple external corpora. Our model achieves high performance and strong generalization across topical domains. We illustrate two applications with the fine-tuned model: (1) the improvement in persuasiveness prediction from incorporating rhetorical strategy labels, and (2) analyzing temporal and partisan shifts in rhetorical strategies in U.S. Presidential debates (1960-2020), revealing increased use of affective over cognitive argument in U.S. Presidential debates.

cs.CL↗

arXiv2025

Cross-Modality Learning for Predicting IHC Biomarkers from H&E-Stained Whole-Slide Images

Hematoxylin and Eosin (H&E) staining is a cornerstone of pathological analysis, offering reliable visualization of cellular morphology and tissue architecture for cancer diagnosis, subtyping, and grading. Immunohistochemistry (IHC) staining provides molecular insights by detecting specific proteins within tissues, enhancing diagnostic accuracy, and improving treatment planning. However, IHC staining is costly, time-consuming, and resource-intensive, requiring specialized expertise. To address these limitations, this study proposes HistoStainAlign, a novel deep learning framework that predicts IHC staining patterns directly from H&E whole-slide images (WSIs) by learning joint representations of morphological and molecular features. The framework integrates paired H&E and IHC embeddings through a contrastive training strategy, capturing complementary features across staining modalities without patch-level annotations or tissue registration. The model was evaluated on gastrointestinal and lung tissue WSIs with three commonly used IHC stains: P53, PD-L1, and Ki-67. HistoStainAlign achieved weighted F1 scores of 0.735 [95% Confidence Interval (CI): 0.670-0.799], 0.830 [95% CI: 0.772-0.886], and 0.723 [95% CI: 0.607-0.836], respectively for these three IHC stains. Embedding analyses demonstrated the robustness of the contrastive alignment in capturing meaningful cross-stain relationships. Comparisons with a baseline model further highlight the advantage of incorporating contrastive learning for improved stain pattern prediction. This study demonstrates the potential of computational approaches to serve as a pre-screening tool, helping prioritize cases for IHC staining and improving workflow efficiency.

eess.IV↗

arXiv2025

Improving Colorectal Cancer Screening and Risk Assessment through Predictive Modeling on Medical Images and Records

Colonoscopy screening effectively identifies and removes polyps before they progress to colorectal cancer (CRC), but current follow-up guidelines rely primarily on histopathological features, overlooking other important CRC risk factors. Variability in polyp characterization among pathologists also hinders consistent surveillance decisions. Advances in digital pathology and deep learning enable the integration of pathology slides and medical records for more accurate CRC risk prediction. Using data from the New Hampshire Colonoscopy Registry, including longitudinal follow-up, we adapted a transformer-based model for histopathology image analysis to predict 5-year CRC risk. We further explored multi-modal fusion strategies to combine clinical records with deep learning-derived image features. Training the model to predict intermediate clinical variables improved 5-year CRC risk prediction (AUC = 0.630) compared to direct prediction (AUC = 0.615, p = 0.013). Incorporating both imaging and non-imaging data, without requiring manual slide review, further improved performance (AUC = 0.674) compared to traditional features from colonoscopy and microscopy reports (AUC = 0.655, p = 0.001). These results highlight the value of integrating diverse data modalities with computational methods to enhance CRC risk stratification.

cs.CV↗

arXiv2025

Transformer-Based Representation Learning for Robust Gene Expression Modeling and Cancer Prognosis

Transformer-based models have achieved remarkable success in natural language and vision tasks, but their application to gene expression analysis remains limited due to data sparsity, high dimensionality, and missing values. We present GexBERT, a transformer-based autoencoder framework for robust representation learning of gene expression data. GexBERT learns context-aware gene embeddings by pretraining on large-scale transcriptomic profiles with a masking and restoration objective that captures co-expression relationships among thousands of genes. We evaluate GexBERT across three critical tasks in cancer research: pan-cancer classification, cancer-specific survival prediction, and missing value imputation. GexBERT achieves state-of-the-art classification accuracy from limited gene subsets, improves survival prediction by restoring expression of prognostic anchor genes, and outperforms conventional imputation methods under high missingness. Furthermore, its attention-based interpretability reveals biologically meaningful gene patterns across cancer types. These findings demonstrate the utility of GexBERT as a scalable and effective tool for gene expression modeling, with translational potential in settings where gene coverage is limited or incomplete.

cs.LG↗

arXiv2025

Predicting Targeted Therapy Resistance in Non-Small Cell Lung Cancer Using Multimodal Machine Learning

Lung cancer is the primary cause of cancer death globally, with non-small cell lung cancer (NSCLC) emerging as its most prevalent subtype. Among NSCLC patients, approximately 32.3% have mutations in the epidermal growth factor receptor (EGFR) gene. Osimertinib, a third-generation EGFR-tyrosine kinase inhibitor (TKI), has demonstrated remarkable efficacy in the treatment of NSCLC patients with activating and T790M resistance EGFR mutations. Despite its established efficacy, drug resistance poses a significant challenge for patients to fully benefit from osimertinib. The absence of a standard tool to accurately predict TKI resistance, including that of osimertinib, remains a critical obstacle. To bridge this gap, in this study, we developed an interpretable multimodal machine learning model designed to predict patient resistance to osimertinib among late-stage NSCLC patients with activating EGFR mutations, achieving a c-index of 0.82 on a multi-institutional dataset. This machine learning model harnesses readily available data routinely collected during patient visits and medical assessments to facilitate precision lung cancer management and informed treatment decisions. By integrating various data types such as histology images, next generation sequencing (NGS) data, demographics data, and clinical records, our multimodal model can generate well-informed recommendations. Our experiment results also demonstrated the superior performance of the multimodal model over single modality models (c-index 0.82 compared with 0.75 and 0.77), thus underscoring the benefit of combining multiple modalities in patient outcome prediction.

cs.LG↗

arXiv2025

ImpScore: A Learnable Metric For Quantifying The Implicitness Level of Sentence

Handling implicit language is essential for natural language processing systems to achieve precise text understanding and facilitate natural interactions with users. Despite its importance, the absence of a metric for accurately measuring the implicitness of language significantly constrains the depth of analysis possible in evaluating models' comprehension capabilities. This paper addresses this gap by developing a scalar metric that quantifies the implicitness level of language without relying on external references. Drawing on principles from traditional linguistics, we define "implicitness" as the divergence between semantic meaning and pragmatic interpretation. To operationalize this definition, we introduce ImpScore, a reference-free metric formulated through an interpretable regression model. This model is trained using pairwise contrastive learning on a specially curated dataset consisting of (implicit sentence, explicit sentence) pairs. We validate ImpScore through a user study that compares its assessments with human evaluations on out-of-distribution data, demonstrating its accuracy and strong correlation with human judgments. Additionally, we apply ImpScore to hate speech detection datasets, illustrating its utility and highlighting significant limitations in current large language models' ability to understand highly implicit content. Our metric is publicly available at https://github.com/audreycs/ImpScore.

cs.CL↗

arXiv2025

Communication is All You Need: Persuasion Dataset Construction via Multi-LLM Communication

Large Language Models (LLMs) have shown proficiency in generating persuasive dialogue, yet concerns about the fluency and sophistication of their outputs persist. This paper presents a multi-LLM communication framework designed to enhance the generation of persuasive data automatically. This framework facilitates the efficient production of high-quality, diverse linguistic content with minimal human oversight. Through extensive evaluations, we demonstrate that the generated data excels in naturalness, linguistic diversity, and the strategic use of persuasion, even in complex scenarios involving social taboos. The framework also proves adept at generalizing across novel contexts. Our results highlight the framework's potential to significantly advance research in both computational and social science domains concerning persuasive communication.

cs.CL↗

arXiv2024

A Benchmark for Long-Form Medical Question Answering

There is a lack of benchmarks for evaluating large language models (LLMs) in long-form medical question answering (QA). Most existing medical QA evaluation benchmarks focus on automatic metrics and multiple-choice questions. While valuable, these benchmarks fail to fully capture or assess the complexities of real-world clinical applications where LLMs are being deployed. Furthermore, existing studies on evaluating long-form answer generation in medical QA are primarily closed-source, lacking access to human medical expert annotations, which makes it difficult to reproduce results and enhance existing baselines. In this work, we introduce a new publicly available benchmark featuring real-world consumer medical questions with long-form answer evaluations annotated by medical doctors. We performed pairwise comparisons of responses from various open and closed-source medical and general-purpose LLMs based on criteria such as correctness, helpfulness, harmfulness, and bias. Additionally, we performed a comprehensive LLM-as-a-judge analysis to study the alignment between human judgments and LLMs. Our preliminary results highlight the strong potential of open LLMs in medical QA compared to leading closed models. Code & Data: https://github.com/lavita-ai/medical-eval-sphere

cs.CL↗

arXiv2024

Ethical Challenges and Evolving Strategies in the Integration of Artificial Intelligence into Clinical Practice

Artificial intelligence (AI) has rapidly transformed various sectors, including healthcare, where it holds the potential to revolutionize clinical practice and improve patient outcomes. However, its integration into medical settings brings significant ethical challenges that need careful consideration. This paper examines the current state of AI in healthcare, focusing on five critical ethical concerns: justice and fairness, transparency, patient consent and confidentiality, accountability, and patient-centered and equitable care. These concerns are particularly pressing as AI systems can perpetuate or even exacerbate existing biases, often resulting from non-representative datasets and opaque model development processes. The paper explores how bias, lack of transparency, and challenges in maintaining patient trust can undermine the effectiveness and fairness of AI applications in healthcare. In addition, we review existing frameworks for the regulation and deployment of AI, identifying gaps that limit the widespread adoption of these systems in a just and equitable manner. Our analysis provides recommendations to address these ethical challenges, emphasizing the need for fairness in algorithm design, transparency in model decision-making, and patient-centered approaches to consent and data privacy. By highlighting the importance of continuous ethical scrutiny and collaboration between AI developers, clinicians, and ethicists, we outline pathways for achieving more responsible and inclusive AI implementation in healthcare. These strategies, if adopted, could enhance both the clinical value of AI and the trustworthiness of AI systems among patients and healthcare professionals, ensuring that these technologies serve all populations equitably.

cs.CY↗

arXiv2024

Deep Learning for Classification of Inflammatory Bowel Disease Activity in Whole Slide Images of Colonic Histopathology

Grading inflammatory bowel disease (IBD) activity using standardized histopathological scoring systems remains challenging due to resource constraints and inter-observer variability. In this study, we developed a deep learning model to classify activity grades in hematoxylin and eosin-stained whole slide images (WSIs) from patients with IBD, offering a robust approach for general pathologists. We utilized 2,077 WSIs from 636 patients treated at Dartmouth-Hitchcock Medical Center in 2018 and 2019, scanned at 40x magnification (0.25 micron/pixel). Board-certified gastrointestinal pathologists categorized the WSIs into four activity classes: inactive, mildly active, moderately active, and severely active. A transformer-based model was developed and validated using five-fold cross-validation to classify IBD activity. Using HoVerNet, we examined neutrophil distribution across activity grades. Attention maps from our model highlighted areas contributing to its prediction. The model classified IBD activity with weighted averages of 0.871 [95% Confidence Interval (CI): 0.860-0.883] for the area under the curve, 0.695 [95% CI: 0.674-0.715] for precision, 0.697 [95% CI: 0.678-0.716] for recall, and 0.695 [95% CI: 0.674-0.714] for F1-score. Neutrophil distribution was significantly different across activity classes. Qualitative evaluation of attention maps by a gastrointestinal pathologist suggested their potential for improved interpretability. Our model demonstrates robust diagnostic performance and could enhance consistency and efficiency in IBD activity assessment.

cs.CV↗

arXiv2024

A Novel Framework for the Automated Characterization of Gram-Stained Blood Culture Slides Using a Large-Scale Vision Transformer

This study introduces a new framework for the artificial intelligence-assisted characterization of Gram-stained whole-slide images (WSIs). As a test for the diagnosis of bloodstream infections, Gram stains provide critical early data to inform patient treatment. Rapid and reliable analysis of Gram stains has been shown to be positively associated with better clinical outcomes, underscoring the need for improved tools to automate Gram stain analysis. In this work, we developed a novel transformer-based model for Gram-stained WSI classification, which is more scalable to large datasets than previous convolutional neural network (CNN) -based methods as it does not require patch-level manual annotations. We also introduce a large Gram stain dataset from Dartmouth-Hitchcock Medical Center (Lebanon, New Hampshire, USA) to evaluate our model, exploring the classification of five major categories of Gram-stained WSIs: Gram-positive cocci in clusters, Gram-positive cocci in pairs/chains, Gram-positive rods, Gram-negative rods, and slides with no bacteria. Our model achieves a classification accuracy of 0.858 (95% CI: 0.805, 0.905) and an AUC of 0.952 (95% CI: 0.922, 0.976) using five-fold nested cross-validation on our 475-slide dataset, demonstrating the potential of large-scale transformer models for Gram stain classification. We further demonstrate the generalizability of our trained model, which achieves strong performance on external datasets without additional fine-tuning.

eess.IV↗

arXiv2024

MentalManip: A Dataset For Fine-grained Analysis of Mental Manipulation in Conversations

Mental manipulation, a significant form of abuse in interpersonal conversations, presents a challenge to identify due to its context-dependent and often subtle nature. The detection of manipulative language is essential for protecting potential victims, yet the field of Natural Language Processing (NLP) currently faces a scarcity of resources and research on this topic. Our study addresses this gap by introducing a new dataset, named ${\rm M{\small ental}M{\small anip}}$, which consists of $4,000$ annotated movie dialogues. This dataset enables a comprehensive analysis of mental manipulation, pinpointing both the techniques utilized for manipulation and the vulnerabilities targeted in victims. Our research further explores the effectiveness of leading-edge models in recognizing manipulative dialogue and its components through a series of experiments with various configurations. The results demonstrate that these models inadequately identify and categorize manipulative content. Attempts to improve their performance by fine-tuning with existing datasets on mental health and toxicity have not overcome these limitations. We anticipate that ${\rm M{\small ental}M{\small anip}}$ will stimulate further research, leading to progress in both understanding and mitigating the impact of mental manipulation in conversations.

cs.CL↗

arXiv2024

Vision Transformer-Based Deep Learning for Histologic Classification of Endometrial Cancer

Endometrial cancer, the fourth most common cancer in females in the United States, with the lifetime risk for developing this disease is approximately 2.8% in women. Precise histologic evaluation and molecular classification of endometrial cancer is important for effective patient management and determining the best treatment modalities. This study introduces EndoNet, which uses convolutional neural networks for extracting histologic features and a vision transformer for aggregating these features and classifying slides based on their visual characteristics into high- and low- grade. The model was trained on 929 digitized hematoxylin and eosin-stained whole-slide images of endometrial cancer from hysterectomy cases at Dartmouth-Health. It classifies these slides into low-grade (Endometroid Grades 1 and 2) and high-grade (endometroid carcinoma FIGO grade 3, uterine serous carcinoma, carcinosarcoma) categories. EndoNet was evaluated on an internal test set of 110 patients and an external test set of 100 patients from the public TCGA database. The model achieved a weighted average F1-score of 0.91 (95% CI: 0.86-0.95) and an AUC of 0.95 (95% CI: 0.89-0.99) on the internal test, and 0.86 (95% CI: 0.80-0.94) for F1-score and 0.86 (95% CI: 0.75-0.93) for AUC on the external test. Pending further validation, EndoNet has the potential to support pathologists without the need of manual annotations in classifying the grades of gynecologic pathology tumors.

cs.CV↗

arXiv2024

Towards Interpretable Deep Reinforcement Learning Models via Inverse Reinforcement Learning

Artificial intelligence, particularly through recent advancements in deep learning, has achieved exceptional performances in many tasks in fields such as natural language processing and computer vision. In addition to desirable evaluation metrics, a high level of interpretability is often required for these models to be reliably utilized. Therefore, explanations that offer insight into the process by which a model maps its inputs onto its outputs are much sought-after. Unfortunately, the current black box nature of machine learning models is still an unresolved issue and this very nature prevents researchers from learning and providing explicative descriptions for a model's behavior and final predictions. In this work, we propose a novel framework utilizing Adversarial Inverse Reinforcement Learning that can provide global explanations for decisions made by a Reinforcement Learning model and capture intuitive tendencies that the model follows by summarizing the model's decision-making process.

cs.LG↗