arXiv subjects

Zihao Fu

Publications and source records attributed to Zihao Fu.

At least 19 recordsLinked to original sources

arXiv2026

The Heterogeneous Safety Impacts of Benign Multilingual Fine-Tuning

Fine-tuning a large language model is a ubiquitous method for enhancing its capability on a specific downstream task. However, prior work has shown that this increase in capability comes with a cost: it can increase a model's tendency to respond to unsafe adversarial prompts, even when fine-tuning with non-adversarial data. We present the first comprehensive empirical study of this phenomenon in multilingual settings by fine-tuning Llama-3.2, Qwen3, and Gemma-3 models using benign data translated across nine languages. We find that safety outcomes are highly sensitive to both the choice of fine-tuning language and the evaluation language, with adversarial compliance rates increasing four-fold in some settings. Multilingual safety drift is decoupled from general capability metrics, and occurs heterogeneously across languages and models. Fine-tuning in non-English languages often induces smaller internal representational drifts than English, but these shifts lead models to default to either exaggerated compliance or refusal. As such, assessing fine-tuning impacts solely in English provides inadequate assurance for deployment. To facilitate further research into these cross-lingual safety blind spots, we release the Multilingual-Benign-Tune dataset and the SORRY-Bench-Multilingual evaluation suite.

cs.CL↗

arXiv2026

Running the Gauntlet: Re-evaluating the Capabilities of Agents Beyond Familiar Environments

As agentic systems continue to evolve and are widely deployed in real-world scenarios, there is a growing demand to faithfully evaluate their capabilities. However, current benchmarks are typically built on popular applications with relatively simple tasks and focus on a narrow set of capabilities while overlooking broader dimensions, resulting in saturated performance on modern agents and failing to probe their limitations. To this end, we introduce GauntletBench, a web-based benchmark for evaluating agent generalisation in challenging scenarios, focusing on three underexplored capabilities (temporal perception, graphical understanding, and 3D reasoning), across five less-covered professional applications (Video Editor, Workflow Builder, 3D Modeller, Flight Analyser, and Circuit Designer), each with 27 vision-intensive tasks (135 in total). Our benchmark provides a modular pipeline that comprises an environment compatible with both open- and closed-source agent frameworks, a controlled web-based application, a well-structured task suite, and an automated evaluation engine with diverse metrics. Contrary to widespread expectations, our empirical results reveal that frontier agentic systems remain far from achieving human-level performance. Even the state-of-the-art agent achieves only a 28.2% success rate on our GauntletBench, highlighting the limitations in these overlooked capabilities and generalisation. By comparison, non-expert human annotators achieve over 80% success on our challenging yet feasible tasks, revealing the substantial gap between current agent capabilities and those required for complex real-world scenarios.

cs.LG↗

arXiv2026

Grounding Text Embeddings in Stakeholder Associations

Text embeddings are widely used to analyse large corpora of complex texts. However, it is unclear whether the embeddings capture the same semantic distances as the human experts using them. Ensuring alignment between embedding representations and human intentions is essential for valid analyses. We present the Stakeholder Grounding Exercise, a method for making expert associations explicit and grounding embedding model results in human understanding. In our primary case study on Danish policy issues, we find that neural text embeddings are substantially less reliable than human experts (19-26 pp gap), and that this misalignment propagates to downstream clustering performance (Spearman $\rho=0.9$ between exercise ranking and cluster quality). A secondary study on US Federal AI use cases replicates the gap (16pp) in English, using a digital protocol and a different community of experts -- demonstrating that the gap is not an artefact of a single instrument or domain. The Stakeholder Grounding Exercise offers a practical method for assessing whether embedding models capture the semantic distinctions that matter most to domain experts.

cs.CL↗

arXiv2026

Flexi-LoRA with Input-Adaptive Ranks: Efficient Finetuning for Speech and Reasoning Tasks

Parameter-efficient fine-tuning methods like Low-Rank Adaptation (LoRA) have become essential for deploying large language models, yet their static parameter allocation remains suboptimal for inputs of varying complexity. We present Flexi-LoRA, a novel framework that dynamically adjusts LoRA ranks based on input complexity during both training and inference. Through empirical analysis across question answering, mathematical reasoning, and speech tasks, we demonstrate that maintaining consistency between training and inference dynamics is important for effective adaptation, particularly for sequential reasoning tasks. Our findings reveal that input-dependent parameter allocation achieves higher performance with fewer parameters by optimally matching rank configurations to question complexity. Furthermore, task-specific dependency on rank dynamics varies, with mathematical reasoning tasks exhibiting higher dependency than QA tasks. Successful adaptation manifests not only in correctness but also in reasoning quality and instruction adherence. Flexi-LoRA consistently outperforms static LoRA while using fewer parameters, with performance gains more pronounced on tasks requiring strict reasoning chains. Our approach realizes key benefits of mixture-of-experts frameworks through a more streamlined implementation, reducing parameter redundancy while improving model capabilities. We provide comprehensive empirical studies across diverse tasks, establishing a basis for future work in input-adaptive and efficient fine-tuning approaches.

cs.LG↗

arXiv2026

Evaluating the Ability of Explanations to Disambiguate Models in a Rashomon Set

Explainable artificial intelligence (XAI) is concerned with producing explanations indicating the inner workings of models. For a Rashomon set of similarly performing models, explanations provide a way of disambiguating the behavior of individual models, helping select models for deployment. However explanations themselves can vary depending on the explainer used, and need to be evaluated. In the paper "Evaluating Model Explanations without Ground Truth", we proposed three principles of explanation evaluation and a new method "AXE" to evaluate the quality of feature-importance explanations. We go on to illustrate how evaluation metrics that rely on comparing model explanations against ideal ground truth explanations obscure behavioral differences within a Rashomon set. Explanation evaluation aligned with our proposed principles would highlight these differences instead, helping select models from the Rashomon set. The selection of alternate models from the Rashomon set can maintain identical predictions but mislead explainers into generating false explanations, and mislead evaluation methods into considering the false explanations to be of high quality. AXE, our proposed explanation evaluation method, can detect this adversarial fairwashing of explanations with a 100% success rate. Unlike prior explanation evaluation strategies such as those based on model sensitivity or ground truth comparison, AXE can determine when protected attributes are used to make predictions.

cs.AI↗

arXiv2026

SCALPEL: Selective Capability Ablation via Low-rank Parameter Editing for Large Language Model Interpretability Analysis

Large language models excel across diverse domains, yet their deployment in healthcare, legal systems, and autonomous decision-making remains limited by incomplete understanding of their internal mechanisms. As these models integrate into high-stakes systems, understanding how they encode capabilities has become fundamental to interpretability research. Traditional approaches identify important modules through gradient attribution or activation analysis, assuming specific capabilities map to specific components. However, this oversimplifies neural computation: modules may contribute to multiple capabilities simultaneously, while single capabilities may distribute across multiple modules. These coarse-grained analyses fail to capture fine-grained, distributed capability encoding. We present SCALPEL (Selective Capability Ablation via Low-rank Parameter Editing for Large language models), a framework representing capabilities as low-rank parameter subspaces rather than discrete modules. Our key insight is that capabilities can be characterized by low-rank modifications distributed across layers and modules, enabling precise capability removal without affecting others. By training LoRA adapters to reduce distinguishing correct from incorrect answers while preserving general language modeling quality, SCALPEL identifies low-rank representations responsible for particular capabilities while remaining disentangled from others. Experiments across diverse capability and linguistic tasks from BLiMP demonstrate that SCALPEL successfully removes target capabilities while preserving general capabilities, providing fine-grained insights into capability distribution across parameter space. Results reveal that capabilities exhibit low-rank structure and can be selectively ablated through targeted parameter-space interventions, offering nuanced understanding of capability encoding in LLMs.

cs.LG↗

arXiv2025

OxEnsemble: Fair Ensembles for Low-Data Classification

We address the problem of fair classification in settings where data is scarce and unbalanced across demographic groups. Such low-data regimes are common in domains like medical imaging, where false negatives can have fatal consequences. We propose a novel approach \emph{OxEnsemble} for efficiently training ensembles and enforcing fairness in these low-data regimes. Unlike other approaches, we aggregate predictions across ensemble members, each trained to satisfy fairness constraints. By construction, \emph{OxEnsemble} is both data-efficient -- carefully reusing held-out data to enforce fairness reliably -- and compute-efficient, requiring little more compute than used to fine-tune or evaluate an existing model. We validate this approach with new theoretical guarantees. Experimentally, our approach yields more consistent outcomes and stronger fairness-accuracy trade-offs than existing methods across multiple challenging medical imaging classification datasets.

cs.CV↗

arXiv2025

Ripple-assisted adsorption of noble gases on graphene at room temperature

Controllable gas adsorption is critical for both scientific and industrial fields, and high-capacity adsorption of gases on solid surfaces provides a significant promise due to its high-safety and low-energy consumption. However, the adsorption of nonpolar gases, particularly noble gases, poses a considerable challenge under atmospheric pressure and room temperature (RT). Here, we theoretically simulate and experimentally realize the stable adsorption of noble gases like xenon (Xe), krypton (Kr), argon (Ar), and helium (He) on highly rippled graphene at RT. The elemental characteristics of adsorbed Xe are confirmed by electron energy loss spectroscopy and X-ray photoelectron spectroscopy. The adsorbed gas atoms are crystalized with periodic arrangements. These adsorbed noble gases on graphene exhibit high stability at RT and can be completely desorbed at approximately 350 {\deg}C without damaging the intrinsic lattice of graphene. The structural and physical properties of graphene are significantly influenced by the adsorbed gas, and they fully recover after desorption. Additionally, this controllable adsorption could be generalized to other layered adsorbents such as NbSe2, MoS2 and carbon nanotubes. We anticipate that this ripple-assisted adsorption will not only re-define the theoretical framework of gas adsorption, but also accelerate advancements in gas storage and separation technologies, as well as enhance the applications in catalysis, surface modification, and other related fields.

cond-mat.mtrl-sci↗

arXiv2025

FairImagen: Post-Processing for Bias Mitigation in Text-to-Image Models

Text-to-image diffusion models, such as Stable Diffusion, have demonstrated remarkable capabilities in generating high-quality and diverse images from natural language prompts. However, recent studies reveal that these models often replicate and amplify societal biases, particularly along demographic attributes like gender and race. In this paper, we introduce FairImagen (https://github.com/fuzihaofzh/FairImagen), a post-hoc debiasing framework that operates on prompt embeddings to mitigate such biases without retraining or modifying the underlying diffusion model. Our method integrates Fair Principal Component Analysis to project CLIP-based input embeddings into a subspace that minimizes group-specific information while preserving semantic content. We further enhance debiasing effectiveness through empirical noise injection and propose a unified cross-demographic projection method that enables simultaneous debiasing across multiple demographic attributes. Extensive experiments across gender, race, and intersectional settings demonstrate that FairImagen significantly improves fairness with a moderate trade-off in image quality and prompt fidelity. Our framework outperforms existing post-hoc methods and offers a simple, scalable, and model-agnostic solution for equitable text-to-image generation.

cs.LG↗

arXiv2025

CAST: Compositional Analysis via Spectral Tracking for Understanding Transformer Layer Functions

Large language models have achieved remarkable success but remain largely black boxes with poorly understood internal mechanisms. To address this limitation, many researchers have proposed various interpretability methods including mechanistic analysis, probing classifiers, and activation visualization, each providing valuable insights from different perspectives. Building upon this rich landscape of complementary approaches, we introduce CAST (Compositional Analysis via Spectral Tracking), a probe-free framework that contributes a novel perspective by analyzing transformer layer functions through direct transformation matrix estimation and comprehensive spectral analysis. CAST offers complementary insights to existing methods by estimating the realized transformation matrices for each layer using Moore-Penrose pseudoinverse and applying spectral analysis with six interpretable metrics characterizing layer behavior. Our analysis reveals distinct behaviors between encoder-only and decoder-only models, with decoder models exhibiting compression-expansion cycles while encoder models maintain consistent high-rank processing. Kernel analysis further demonstrates functional relationship patterns between layers, with CKA similarity matrices clearly partitioning layers into three phases: feature extraction, compression, and specialization.

cs.LG↗

arXiv2025

Multi-use LLM Watermarking and the False Detection Problem

Digital watermarking is a promising solution for mitigating some of the risks arising from the misuse of automatically generated text. These approaches either embed non-specific watermarks to allow for the detection of any text generated by a particular sampler, or embed specific keys that allow the identification of the LLM user. However, simultaneously using the same embedding for both detection and user identification leads to a false detection problem, whereby, as user capacity grows, unwatermarked text is increasingly likely to be falsely detected as watermarked. Through theoretical analysis, we identify the underlying causes of this phenomenon. Building on these insights, we propose Dual Watermarking which jointly encodes detection and identification watermarks into generated text, significantly reducing false positives while maintaining high detection accuracy. Our experimental results validate our theoretical findings and demonstrate the effectiveness of our approach.

cs.CR↗

arXiv2025

Evaluating Model Explanations without Ground Truth

There can be many competing and contradictory explanations for a single model prediction, making it difficult to select which one to use. Current explanation evaluation frameworks measure quality by comparing against ideal "ground-truth" explanations, or by verifying model sensitivity to important inputs. We outline the limitations of these approaches, and propose three desirable principles to ground the future development of explanation evaluation strategies for local feature importance explanations. We propose a ground-truth Agnostic eXplanation Evaluation framework (AXE) for evaluating and comparing model explanations that satisfies these principles. Unlike prior approaches, AXE does not require access to ideal ground-truth explanations for comparison, or rely on model sensitivity - providing an independent measure of explanation quality. We verify AXE by comparing with baselines, and show how it can be used to detect explanation fairwashing. Our code is available at https://github.com/KaiRawal/Evaluating-Model-Explanations-without-Ground-Truth.

cs.AI↗

arXiv2025

CLIP-SENet: CLIP-based Semantic Enhancement Network for Vehicle Re-identification

Vehicle re-identification (Re-ID) is a crucial task in intelligent transportation systems (ITS), aimed at retrieving and matching the same vehicle across different surveillance cameras. Numerous studies have explored methods to enhance vehicle Re-ID by focusing on semantic enhancement. However, these methods often rely on additional annotated information to enable models to extract effective semantic features, which brings many limitations. In this work, we propose a CLIP-based Semantic Enhancement Network (CLIP-SENet), an end-to-end framework designed to autonomously extract and refine vehicle semantic attributes, facilitating the generation of more robust semantic feature representations. Inspired by zero-shot solutions for downstream tasks presented by large-scale vision-language models, we leverage the powerful cross-modal descriptive capabilities of the CLIP image encoder to initially extract general semantic information. Instead of using a text encoder for semantic alignment, we design an adaptive fine-grained enhancement module (AFEM) to adaptively enhance this general semantic information at a fine-grained level to obtain robust semantic feature representations. These features are then fused with common Re-ID appearance features to further refine the distinctions between vehicles. Our comprehensive evaluation on three benchmark datasets demonstrates the effectiveness of CLIP-SENet. Our approach achieves new state-of-the-art performance, with 92.9% mAP and 98.7% Rank-1 on VeRi-776 dataset, 90.4% Rank-1 and 98.7% Rank-5 on VehicleID dataset, and 89.1% mAP and 97.9% Rank-1 on the more challenging VeRi-Wild dataset.

cs.CV↗

arXiv2024

OxonFair: A Flexible Toolkit for Algorithmic Fairness

We present OxonFair, a new open source toolkit for enforcing fairness in binary classification. Compared to existing toolkits: (i) We support NLP and Computer Vision classification as well as standard tabular problems. (ii) We support enforcing fairness on validation data, making us robust to a wide range of overfitting challenges. (iii) Our approach can optimize any measure based on True Positives, False Positive, False Negatives, and True Negatives. This makes it easily extensible and much more expressive than existing toolkits. It supports all 9 and all 10 of the decision-based group metrics of two popular review articles. (iv) We jointly optimize a performance objective alongside fairness constraints. This minimizes degradation while enforcing fairness, and even improves the performance of inadequately tuned unfair baselines. OxonFair is compatible with standard ML toolkits, including sklearn, Autogluon, and PyTorch and is available at https://github.com/oxfordinternetinstitute/oxonfair

cs.CY↗

arXiv2024

Large Language Model for Table Processing: A Survey

Tables, typically two-dimensional and structured to store large amounts of data, are essential in daily activities like database queries, spreadsheet manipulations, web table question answering, and image table information extraction. Automating these table-centric tasks with Large Language Models (LLMs) or Visual Language Models (VLMs) offers significant public benefits, garnering interest from academia and industry. This survey provides a comprehensive overview of table-related tasks, examining both user scenarios and technical aspects. It covers traditional tasks like table question answering as well as emerging fields such as spreadsheet manipulation and table data analysis. We summarize the training techniques for LLMs and VLMs tailored for table processing. Additionally, we discuss prompt engineering, particularly the use of LLM-powered agents, for various table-related tasks. Finally, we highlight several challenges, including diverse user input when serving and slow thinking using chain-of-thought.

cs.AI↗

arXiv2023

Repetition In Repetition Out: Towards Understanding Neural Text Degeneration from the Data Perspective

There are a number of diverging hypotheses about the neural text degeneration problem, i.e., generating repetitive and dull loops, which makes this problem both interesting and confusing. In this work, we aim to advance our understanding by presenting a straightforward and fundamental explanation from the data perspective. Our preliminary investigation reveals a strong correlation between the degeneration issue and the presence of repetitions in training data. Subsequent experiments also demonstrate that by selectively dropping out the attention to repetitive words in training data, degeneration can be significantly minimized. Furthermore, our empirical analysis illustrates that prior works addressing the degeneration issue from various standpoints, such as the high-inflow words, the likelihood objective, and the self-reinforcement phenomenon, can be interpreted by one simple explanation. That is, penalizing the repetitions in training data is a common and fundamental factor for their effectiveness. Moreover, our experiments reveal that penalizing the repetitions in training data remains critical even when considering larger model sizes and instruction tuning.

cs.CL↗

arXiv2023

BAND: Biomedical Alert News Dataset

Infectious disease outbreaks continue to pose a significant threat to human health and well-being. To improve disease surveillance and understanding of disease spread, several surveillance systems have been developed to monitor daily news alerts and social media. However, existing systems lack thorough epidemiological analysis in relation to corresponding alerts or news, largely due to the scarcity of well-annotated reports data. To address this gap, we introduce the Biomedical Alert News Dataset (BAND), which includes 1,508 samples from existing reported news articles, open emails, and alerts, as well as 30 epidemiology-related questions. These questions necessitate the model's expert reasoning abilities, thereby offering valuable insights into the outbreak of the disease. The BAND dataset brings new challenges to the NLP world, requiring better disguise capability of the content and the ability to infer important information. We provide several benchmark tasks, including Named Entity Recognition (NER), Question Answering (QA), and Event Extraction (EE), to show how existing models are capable of handling these tasks in the epidemiology domain. To the best of our knowledge, the BAND corpus is the largest corpus of well-annotated biomedical outbreak alert news with elaborately designed questions, making it a valuable resource for epidemiologists and NLP researchers alike.

cs.CL↗

arXiv2023

Biomedical Named Entity Recognition via Dictionary-based Synonym Generalization

Biomedical named entity recognition is one of the core tasks in biomedical natural language processing (BioNLP). To tackle this task, numerous supervised/distantly supervised approaches have been proposed. Despite their remarkable success, these approaches inescapably demand laborious human effort. To alleviate the need of human effort, dictionary-based approaches have been proposed to extract named entities simply based on a given dictionary. However, one downside of existing dictionary-based approaches is that they are challenged to identify concept synonyms that are not listed in the given dictionary, which we refer as the synonym generalization problem. In this study, we propose a novel Synonym Generalization (SynGen) framework that recognizes the biomedical concepts contained in the input text using span-based predictions. In particular, SynGen introduces two regularization terms, namely, (1) a synonym distance regularizer; and (2) a noise perturbation regularizer, to minimize the synonym generalization error. To demonstrate the effectiveness of our approach, we provide a theoretical analysis of the bound of synonym generalization error. We extensively evaluate our approach on a wide range of benchmarks and the results verify that SynGen outperforms previous dictionary-based models by notable margins. Lastly, we provide a detailed analysis to further reveal the merits and inner-workings of our approach.

cs.CL↗