arXiv subjects

Khai-Nguyen Nguyen

Publications and source records attributed to Khai-Nguyen Nguyen.

12 recordsLinked to original sources

arXiv2026

Deep and shallow biases in language models

Large language models often repeatedly select the same answer even when many alternatives are plausible. Prior work treats this concentration as bias, but it does not distinguish stable model preferences from responses that depend on a particular prompt wording. We introduce a bias depth score that measures both how strongly a model prefers its top answer under direct prompting and whether that answer survives scenario reframing. Across 4,442 opinion prompts and four large language models, only about a quarter of the concentrated preferences survive reframing. We call these persistent cases Deep biases, and the remaining prompt-dependent cases Shallow biases. Our results show that Deep biases are more often inherited from pretraining and preserved through SFT. Under both continued fine-tuning and prompt-based debiasing for diversity, Deep biases are consistently harder to remove than Shallow biases. Bias depth therefore separates stable learned biases from prompt-wording artifacts that single-prompt metrics conflate. Code, models, and data are available at deepbias.github.io.

cs.CL↗

arXiv2026

Pattern over Pixels: Measuring Pattern Completion Bias in Multimodal Code Generation

Multimodal large language models (MLLMs) are increasingly used to translate webpage screenshots into front-end code, but repeated UI patterns may sway them toward visually incorrect yet pattern-consistent outputs. In this work, we test how repeated webpage patterns hurt MLLM accuracy on an objective screenshot-to-code fill-in-the-blank task. We introduce the first benchmark for visual pattern-completion bias, where one localized element in a repeated UI pattern is perturbed and the model must recover the masked width or font-size value from the screenshot and HTML context. Starting from 30 webpages curated from the Design2Code dataset, we build 1,440 evaluated screenshots spanning structural card and text-style patterns under standard and noise-overlaid conditions. We evaluate five frontier MLLMs and find that all are strongly biased toward the repeated baseline. Mean bias rate reaches 69.78% on card-width perturbations and 80.22% on text font-size perturbations, while mean accuracy is only 21.17% and 7.89%, respectively. Codex-5.3 performs best but still drops from 68.61% accuracy on cards to 13.89% on text, while Flash-3.0 reaches 96.11% bias on text. Noise, subtler perturbations, and boundary positions further increase bias rate. Reasoning analysis further shows that greater reasoning effort correlates with lower bias, yet qualitative evidence reveals that models can identify the anomalous element and still override it with the pattern-consistent answer. Our results identify a concrete failure mode in multimodal code generation and show that its severity is strongly associated with visual saliency

cs.SE↗

arXiv2026

Like a bilingual baby: The advantage of visually grounding a bilingual language model

Unlike most neural language models, humans learn language in a rich, multi-sensory and, often, multi-lingual environment. Current language models typically fail to fully capture the complexities of multilingual language use. We train an LSTM language model on images and captions in English and Spanish from MS-COCO-ES. We find that the visual grounding improves the model's understanding of semantic similarity both within and across languages and improves perplexity. However, we find no significant advantage of visual grounding for abstract words. Our results provide additional evidence of the advantages of visually grounded language models and point to the need for more naturalistic language data from multilingual speakers and multilingual datasets with perceptual grounding.

cs.CL↗

arXiv2026

Prior Bias in Vision Language Models on UML Diagram Interpretation

Vision Language Models (VLMs) are increasingly applied to software engineering artifacts, especially UML class diagrams whose meaning depends on visual notation. Yet, it is unclear whether VLMs actually read such diagrams or instead answer from pretrained priors about how classes typically relate. We introduce a controlled UML benchmark in which each prior-conforming diagram is paired with its prior-conflicting counterpart that (1) preserves the same class names and layout while (2) reverses only the relation arrow. We evaluate eight open-source VLMs from two model families, InternVL3.5 and Qwen3, alongside two closed-source frontier models GPT-5.4 and GPT-5.4 Mini. Across the eight open-source models, reversing the arrow reduces relation-direction accuracy by 33.48% on average, while GPT-5.4 Mini retains a 10% gap. In the harder three-class condition, accuracy drops sharply by 45.28% for open-source models, and even 18.62% for the GPT-5.4 family on average. Scaling provides only limited improvements and is family-dependent. Our benchmark presents a diagnostic prior-driven failure in diagram-grounded software understanding. Our artifact is available at https://anonymous.4open.science/r/UMLKnowledgeConflict-8461.

cs.CV↗

arXiv2026

Vision Language Models are Biased

Large language models (LLMs) memorize a vast amount of prior knowledge from the Internet that helps them on downstream tasks but also may notoriously sway their outputs towards wrong or biased answers. In this work, we test how the knowledge about popular subjects hurt the accuracy of vision language models (VLMs) on standard, objective visual tasks of counting and identification. We find that state-of-the-art VLMs are strongly biased (e.g., unable to recognize the 4th stripe has been added to a 3-stripe Adidas logo) scoring an average of 17.05% accuracy in counting (e.g., counting stripes in an Adidas-like logo) across 7 diverse domains from animals, logos, chess, board games, optical illusions, to patterned grids. Removing image backgrounds nearly doubles accuracy (21.09 percentage points), revealing that contextual visual cues trigger these biased responses. Further analysis of VLMs' reasoning patterns shows that counting accuracy initially rises with thinking tokens, reaching ~40%, before declining with excessive reasoning. Our work presents an interesting failure mode in VLMs and a human-supervised automated framework for testing VLM biases. Code and data are available at: vlmsarebiased.github.io.

cs.LG↗

arXiv2025

Toward Explaining Large Language Models in Software Engineering Tasks

Recent progress in Large Language Models (LLMs) has substantially advanced the automation of software engineering (SE) tasks, enabling complex activities such as code generation and code summarization. However, the black-box nature of LLMs remains a major barrier to their adoption in high-stakes and safety-critical domains, where explainability and transparency are vital for trust, accountability, and effective human supervision. Despite increasing interest in explainable AI for software engineering, existing methods lack domain-specific explanations aligned with how practitioners reason about SE artifacts. To address this gap, we introduce FeatureSHAP, the first fully automated, model-agnostic explainability framework tailored to software engineering tasks. Based on Shapley values, FeatureSHAP attributes model outputs to high-level input features through systematic input perturbation and task-specific similarity comparisons, while remaining compatible with both open-source and proprietary LLMs. We evaluate FeatureSHAP on two bi-modal SE tasks: code generation and code summarization. The results show that FeatureSHAP assigns less importance to irrelevant input features and produces explanations with higher fidelity than baseline methods. A practitioner survey involving 37 participants shows that FeatureSHAP helps practitioners better interpret model outputs and make more informed decisions. Collectively, FeatureSHAP represents a meaningful step toward practical explainable AI in software engineering. FeatureSHAP is available at https://github.com/deviserlab/FeatureSHAP.

cs.SE↗

arXiv2025

S-Chain: Structured Visual Chain-of-Thought For Medicine

Faithful reasoning in medical vision-language models (VLMs) requires not only accurate predictions but also transparent alignment between textual rationales and visual evidence. While Chain-of-Thought (CoT) prompting has shown promise in medical visual question answering (VQA), no large-scale expert-level dataset has captured stepwise reasoning with precise visual grounding. We introduce S-Chain, the first large-scale dataset of 12,000 expert-annotated medical images with bounding boxes and structured visual CoT (SV-CoT), explicitly linking visual regions to reasoning steps. The dataset further supports 16 languages, totaling over 700k VQA pairs for broad multilingual applicability. Using S-Chain, we benchmark state-of-the-art medical VLMs (ExGra-Med, LLaVA-Med) and general-purpose VLMs (Qwen2.5-VL, InternVL2.5), showing that SV-CoT supervision significantly improves interpretability, grounding fidelity, and robustness. Beyond benchmarking, we study its synergy with retrieval-augmented generation, revealing how domain knowledge and visual grounding interact during autoregressive reasoning. Finally, we propose a new mechanism that strengthens the alignment between visual evidence and reasoning, improving both reliability and efficiency. S-Chain establishes a new benchmark for grounded medical reasoning and paves the way toward more trustworthy and explainable medical VLMs.

cs.LG↗

arXiv2025

Sentiment Reasoning for Healthcare

Transparency in AI healthcare decision-making is crucial. By incorporating rationales to explain reason for each predicted label, users could understand Large Language Models (LLMs)'s reasoning to make better decision. In this work, we introduce a new task - Sentiment Reasoning - for both speech and text modalities, and our proposed multimodal multitask framework and the world's largest multimodal sentiment analysis dataset. Sentiment Reasoning is an auxiliary task in sentiment analysis where the model predicts both the sentiment label and generates the rationale behind it based on the input transcript. Our study conducted on both human transcripts and Automatic Speech Recognition (ASR) transcripts shows that Sentiment Reasoning helps improve model transparency by providing rationale for model prediction with quality semantically comparable to humans while also improving model's classification performance (+2% increase in both accuracy and macro-F1) via rationale-augmented fine-tuning. Also, no significant difference in the semantic quality of generated rationales between human and ASR transcripts. All code, data (five languages - Vietnamese, English, Chinese, German, and French) and models are published online: https://github.com/leduckhai/Sentiment-Reasoning

cs.CL↗

arXiv2025

Real-time Speech Summarization for Medical Conversations

In doctor-patient conversations, identifying medically relevant information is crucial, posing the need for conversation summarization. In this work, we propose the first deployable real-time speech summarization system for real-world applications in industry, which generates a local summary after every N speech utterances within a conversation and a global summary after the end of a conversation. Our system could enhance user experience from a business standpoint, while also reducing computational costs from a technical perspective. Secondly, we present VietMed-Sum which, to our knowledge, is the first speech summarization dataset for medical conversations. Thirdly, we are the first to utilize LLM and human annotators collaboratively to create gold standard and synthetic summaries for medical conversation summarization. Finally, we present baseline results of state-of-the-art models on VietMed-Sum. All code, data (English-translated and Vietnamese) and models are available online: https://github.com/leduckhai/MultiMed/tree/master/VietMed-Sum

cs.CL↗

arXiv2025

Medical Spoken Named Entity Recognition

Spoken Named Entity Recognition (NER) aims to extract named entities from speech and categorise them into types like person, location, organization, etc. In this work, we present VietMed-NER - the first spoken NER dataset in the medical domain. To our knowledge, our Vietnamese real-world dataset is the largest spoken NER dataset in the world regarding the number of entity types, featuring 18 distinct types. Furthermore, we present baseline results using various state-of-the-art pre-trained models: encoder-only and sequence-to-sequence; and conduct quantitative and qualitative error analysis. We found that pre-trained multilingual models generally outperform monolingual models on reference text and ASR output and encoders outperform sequence-to-sequence models in NER tasks. By translating the transcripts, the dataset can also be utilised for text NER in the medical domain in other languages than Vietnamese. All code, data and models are publicly available: https://github.com/leduckhai/MultiMed/tree/master/VietMed-NER.

eess.AS↗

arXiv2025

Resource-Efficient & Effective Code Summarization

Code Language Models (CLMs) have demonstrated high effectiveness in automating software engineering tasks such as bug fixing, code generation, and code documentation. This progress has been driven by the scaling of large models, ranging from millions to trillions of parameters (e.g., GPT-4). However, as models grow in scale, sustainability concerns emerge, as they are extremely resource-intensive, highlighting the need for efficient, environmentally conscious solutions. GreenAI techniques, such as QLoRA (Quantized Low-Rank Adaptation), offer a promising path for dealing with large models' sustainability as they enable resource-efficient model fine-tuning. Previous research has shown the effectiveness of QLoRA in code-related tasks, particularly those involving natural language inputs and code as the target output (NL-to-Code), such as code generation. However, no studies have explored its application to tasks that are fundamentally similar to NL-to-Code (natural language to code) but operate in the opposite direction, such as code summarization. This leaves a gap in understanding how well QLoRA can generalize to Code-to-NL tasks, which are equally important for supporting developers in understanding and maintaining code. To address this gap, we investigate the extent to which QLoRA's capabilities in NL-to-Code tasks can be leveraged and transferred to code summarization, one representative Code-to-NL task. Our study evaluates two state-of-the-art CLMs (CodeLlama and DeepSeek-Coder) across two programming languages: Python and Java. Our research tasked models with generating descriptions for Python and Java code methods. The results align with prior findings on QLoRA for source code generation, showing that QLoRA enables efficient fine-tuning of CLMs for code summarization.

cs.SE↗

arXiv2023

Getting Away with More Network Pruning: From Sparsity to Geometry and Linear Regions

One surprising trait of neural networks is the extent to which their connections can be pruned with little to no effect on accuracy. But when we cross a critical level of parameter sparsity, pruning any further leads to a sudden drop in accuracy. This drop plausibly reflects a loss in model complexity, which we aim to avoid. In this work, we explore how sparsity also affects the geometry of the linear regions defined by a neural network, and consequently reduces the expected maximum number of linear regions based on the architecture. We observe that pruning affects accuracy similarly to how sparsity affects the number of linear regions and our proposed bound for the maximum number. Conversely, we find out that selecting the sparsity across layers to maximize our bound very often improves accuracy in comparison to pruning as much with the same sparsity in all layers, thereby providing us guidance on where to prune.

cs.LG↗