arXiv subjects

Jiahui Geng

Publications and source records attributed to Jiahui Geng.

36 records · Page 2Linked to original sources

arXiv2025

HD-NDEs: Neural Differential Equations for Hallucination Detection in LLMs

In recent years, large language models (LLMs) have made remarkable advancements, yet hallucination, where models produce inaccurate or non-factual statements, remains a significant challenge for real-world deployment. Although current classification-based methods, such as SAPLMA, are highly efficient in mitigating hallucinations, they struggle when non-factual information arises in the early or mid-sequence of outputs, reducing their reliability. To address these issues, we propose Hallucination Detection-Neural Differential Equations (HD-NDEs), a novel method that systematically assesses the truthfulness of statements by capturing the full dynamics of LLMs within their latent space. Our approaches apply neural differential equations (Neural DEs) to model the dynamic system in the latent space of LLMs. Then, the sequence in the latent space is mapped to the classification space for truth assessment. The extensive experiments across five datasets and six widely used LLMs demonstrate the effectiveness of HD-NDEs, especially, achieving over 14% improvement in AUC-ROC on the True-False dataset compared to state-of-the-art techniques.

cs.CL↗

arXiv2025

VSCBench: Bridging the Gap in Vision-Language Model Safety Calibration

The rapid advancement of vision-language models (VLMs) has brought a lot of attention to their safety alignment. However, existing methods have primarily focused on model undersafety, where the model responds to hazardous queries, while neglecting oversafety, where the model refuses to answer safe queries. In this paper, we introduce the concept of $\textit{safety calibration}$, which systematically addresses both undersafety and oversafety. Specifically, we present $\textbf{VSCBench}$, a novel dataset of 3,600 image-text pairs that are visually or textually similar but differ in terms of safety, which is designed to evaluate safety calibration across image-centric and text-centric scenarios. Based on our benchmark, we evaluate safety calibration across eleven widely used VLMs. Our extensive experiments revealed major issues with both undersafety and oversafety. We further investigated four approaches to improve the model's safety calibration. We found that even though some methods effectively calibrated the models' safety problems, these methods also lead to the degradation of models' utility. This trade-off underscores the urgent need for advanced calibration methods, and our benchmark provides a valuable tool for evaluating future approaches. Our code and data are available at https://github.com/jiahuigeng/VSCBench.git.

cs.IR↗

arXiv2025

Shaping the Safety Boundaries: Understanding and Defending Against Jailbreaks in Large Language Models

Jailbreaking in Large Language Models (LLMs) is a major security concern as it can deceive LLMs to generate harmful text. Yet, there is still insufficient understanding of how jailbreaking works, which makes it hard to develop effective defense strategies. We aim to shed more light into this issue: we conduct a detailed large-scale analysis of seven different jailbreak methods and find that these disagreements stem from insufficient observation samples. In particular, we introduce \textit{safety boundary}, and we find that jailbreaks shift harmful activations outside that safety boundary, where LLMs are less sensitive to harmful information. We also find that the low and the middle layers are critical in such shifts, while deeper layers have less impact. Leveraging on these insights, we propose a novel defense called \textbf{Activation Boundary Defense} (ABD), which adaptively constrains the activations within the safety boundary. We further use Bayesian optimization to selectively apply the defense method to the low and the middle layers. Our experiments on several benchmarks show that ABD achieves an average DSR of over 98\% against various forms of jailbreak attacks, with less than 2\% impact on the model's general capabilities.

cs.CL↗

arXiv2025

SAUCE: Selective Concept Unlearning in Vision-Language Models with Sparse Autoencoders

Unlearning methods for vision-language models (VLMs) have primarily adapted techniques from large language models (LLMs), relying on weight updates that demand extensive annotated forget sets. Moreover, these methods perform unlearning at a coarse granularity, often leading to excessive forgetting and reduced model utility. To address this issue, we introduce SAUCE, a novel method that leverages sparse autoencoders (SAEs) for fine-grained and selective concept unlearning in VLMs. Briefly, SAUCE first trains SAEs to capture high-dimensional, semantically rich sparse features. It then identifies the features most relevant to the target concept for unlearning. During inference, it selectively modifies these features to suppress specific concepts while preserving unrelated information. We evaluate SAUCE on two distinct VLMs, LLaVA-v1.5-7B and LLaMA-3.2-11B-Vision-Instruct, across two types of tasks: concrete concept unlearning (objects and sports scenes) and abstract concept unlearning (emotions, colors, and materials), encompassing a total of 60 concepts. Extensive experiments demonstrate that SAUCE outperforms state-of-the-art methods by 18.04% in unlearning quality while maintaining comparable model utility. Furthermore, we investigate SAUCE's robustness against widely used adversarial attacks, its transferability across models, and its scalability in handling multiple simultaneous unlearning requests. Our findings establish SAUCE as an effective and scalable solution for selective concept unlearning in VLMs.

cs.CV↗

arXiv2025

LLM-DetectAIve: a Tool for Fine-Grained Machine-Generated Text Detection

The ease of access to large language models (LLMs) has enabled a widespread of machine-generated texts, and now it is often hard to tell whether a piece of text was human-written or machine-generated. This raises concerns about potential misuse, particularly within educational and academic domains. Thus, it is important to develop practical systems that can automate the process. Here, we present one such system, LLM-DetectAIve, designed for fine-grained detection. Unlike most previous work on machine-generated text detection, which focused on binary classification, LLM-DetectAIve supports four categories: (i) human-written, (ii) machine-generated, (iii) machine-written, then machine-humanized, and (iv) human-written, then machine-polished. Category (iii) aims to detect attempts to obfuscate the fact that a text was machine-generated, while category (iv) looks for cases where the LLM was used to polish a human-written text, which is typically acceptable in academic writing, but not in education. Our experiments show that LLM-DetectAIve can effectively identify the above four categories, which makes it a potentially useful tool in education, academia, and other domains. LLM-DetectAIve is publicly accessible at https://github.com/mbzuai-nlp/LLM-DetectAIve. The video describing our system is available at https://youtu.be/E8eT_bE7k8c.

cs.CL↗

arXiv2025

GenAI Content Detection Task 1: English and Multilingual Machine-Generated Text Detection: AI vs. Human

We present the GenAI Content Detection Task~1 -- a shared task on binary machine generated text detection, conducted as a part of the GenAI workshop at COLING 2025. The task consists of two subtasks: Monolingual (English) and Multilingual. The shared task attracted many participants: 36 teams made official submissions to the Monolingual subtask during the test phase and 26 teams -- to the Multilingual. We provide a comprehensive overview of the data, a summary of the results -- including system rankings and performance scores -- detailed descriptions of the participating systems, and an in-depth analysis of submissions. https://github.com/mbzuai-nlp/COLING-2025-Workshop-on-MGT-Detection-Task1

cs.CL↗

arXiv2025

FIRE: Fact-checking with Iterative Retrieval and Verification

Fact-checking long-form text is challenging, and it is therefore common practice to break it down into multiple atomic claims. The typical approach to fact-checking these atomic claims involves retrieving a fixed number of pieces of evidence, followed by a verification step. However, this method is usually not cost-effective, as it underutilizes the verification model's internal knowledge of the claim and fails to replicate the iterative reasoning process in human search strategies. To address these limitations, we propose FIRE, a novel agent-based framework that integrates evidence retrieval and claim verification in an iterative manner. Specifically, FIRE employs a unified mechanism to decide whether to provide a final answer or generate a subsequent search query, based on its confidence in the current judgment. We compare FIRE with other strong fact-checking frameworks and find that it achieves slightly better performance while reducing large language model (LLM) costs by an average of 7.6 times and search costs by 16.5 times. These results indicate that FIRE holds promise for application in large-scale fact-checking operations. Our code is available at https://github.com/mbzuai-nlp/fire.git.

cs.IR↗

arXiv2025

Internal Activation Revision: Safeguarding Vision Language Models Without Parameter Update

Vision-language models (VLMs) demonstrate strong multimodal capabilities but have been found to be more susceptible to generating harmful content compared to their backbone large language models (LLMs). Our investigation reveals that the integration of images significantly shifts the model's internal activations during the forward pass, diverging from those triggered by textual input. Moreover, the safety alignments of LLMs embedded within VLMs are not sufficiently robust to handle the activations discrepancies, making the models vulnerable to even the simplest jailbreaking attacks. To address this issue, we propose an \textbf{internal activation revision} approach that efficiently revises activations during generation, steering the model toward safer outputs. Our framework incorporates revisions at both the layer and head levels, offering control over the model's generation at varying levels of granularity. In addition, we explore three strategies for constructing positive and negative samples and two approaches for extracting revision vectors, resulting in different variants of our method. Comprehensive experiments demonstrate that the internal activation revision method significantly improves the safety of widely used VLMs, reducing attack success rates by an average of 48.94\%, 34.34\%, 43.92\%, and 52.98\% on SafeBench, Safe-Unsafe, Unsafe, and MM-SafetyBench, respectively, while minimally impacting model helpfulness.

cs.LG↗

arXiv2024

Reference-free Hallucination Detection for Large Vision-Language Models

Large vision-language models (LVLMs) have made significant progress in recent years. While LVLMs exhibit excellent ability in language understanding, question answering, and conversations of visual inputs, they are prone to producing hallucinations. While several methods are proposed to evaluate the hallucinations in LVLMs, most are reference-based and depend on external tools, which complicates their practical application. To assess the viability of alternative methods, it is critical to understand whether the reference-free approaches, which do not rely on any external tools, can efficiently detect hallucinations. Therefore, we initiate an exploratory study to demonstrate the effectiveness of different reference-free solutions in detecting hallucinations in LVLMs. In particular, we conduct an extensive study on three kinds of techniques: uncertainty-based, consistency-based, and supervised uncertainty quantification methods on four representative LVLMs across two different tasks. The empirical results show that the reference-free approaches are capable of effectively detecting non-factual responses in LVLMs, with the supervised uncertainty quantification method outperforming the others, achieving the best performance across different settings.

cs.CL↗

arXiv2024

CVQA: Culturally-diverse Multilingual Visual Question Answering Benchmark

Visual Question Answering (VQA) is an important task in multimodal AI, and it is often used to test the ability of vision-language models to understand and reason on knowledge present in both visual and textual data. However, most of the current VQA models use datasets that are primarily focused on English and a few major world languages, with images that are typically Western-centric. While recent efforts have tried to increase the number of languages covered on VQA datasets, they still lack diversity in low-resource languages. More importantly, although these datasets often extend their linguistic range via translation or some other approaches, they usually keep images the same, resulting in narrow cultural representation. To address these limitations, we construct CVQA, a new Culturally-diverse multilingual Visual Question Answering benchmark, designed to cover a rich set of languages and cultures, where we engage native speakers and cultural experts in the data collection process. As a result, CVQA includes culturally-driven images and questions from across 30 countries on four continents, covering 31 languages with 13 scripts, providing a total of 10k questions. We then benchmark several Multimodal Large Language Models (MLLMs) on CVQA, and show that the dataset is challenging for the current state-of-the-art models. This benchmark can serve as a probing evaluation suite for assessing the cultural capability and bias of multimodal models and hopefully encourage more research efforts toward increasing cultural awareness and linguistic diversity in this field.

cs.CV↗

arXiv2024

Multimodal Large Language Models to Support Real-World Fact-Checking

Multimodal large language models (MLLMs) carry the potential to support humans in processing vast amounts of information. While MLLMs are already being used as a fact-checking tool, their abilities and limitations in this regard are understudied. Here is aim to bridge this gap. In particular, we propose a framework for systematically assessing the capacity of current multimodal models to facilitate real-world fact-checking. Our methodology is evidence-free, leveraging only these models' intrinsic knowledge and reasoning capabilities. By designing prompts that extract models' predictions, explanations, and confidence levels, we delve into research questions concerning model accuracy, robustness, and reasons for failure. We empirically find that (1) GPT-4V exhibits superior performance in identifying malicious and misleading multimodal claims, with the ability to explain the unreasonable aspects and underlying motives, and (2) existing open-source models exhibit strong biases and are highly sensitive to the prompt. Our study offers insights into combating false multimodal information and building secure, trustworthy multimodal models. To the best of our knowledge, we are the first to evaluate MLLMs for real-world fact-checking.

cs.CL↗

arXiv2024

Factcheck-Bench: Fine-Grained Evaluation Benchmark for Automatic Fact-checkers

The increased use of large language models (LLMs) across a variety of real-world applications calls for mechanisms to verify the factual accuracy of their outputs. In this work, we present a holistic end-to-end solution for annotating the factuality of LLM-generated responses, which encompasses a multi-stage annotation scheme designed to yield detailed labels concerning the verifiability and factual inconsistencies found in LLM outputs. We further construct an open-domain document-level factuality benchmark in three-level granularity: claim, sentence and document, aiming to facilitate the evaluation of automatic fact-checking systems. Preliminary experiments show that FacTool, FactScore and Perplexity.ai are struggling to identify false claims, with the best F1=0.63 by this annotation solution based on GPT-4. Annotation tool, benchmark and code are available at https://github.com/yuxiaw/Factcheck-GPT.

cs.CL↗

arXiv2024

A Survey of Confidence Estimation and Calibration in Large Language Models

Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks in various domains. Despite their impressive performance, they can be unreliable due to factual errors in their generations. Assessing their confidence and calibrating them across different tasks can help mitigate risks and enable LLMs to produce better generations. There has been a lot of recent research aiming to address this, but there has been no comprehensive overview to organize it and outline the main lessons learned. The present survey aims to bridge this gap. In particular, we outline the challenges and we summarize recent technical advancements for LLM confidence estimation and calibration. We further discuss their applications and suggest promising directions for future work.

cs.CL↗

arXiv2023

A Survey on Dataset Distillation: Approaches, Applications and Future Directions

Dataset distillation is attracting more attention in machine learning as training sets continue to grow and the cost of training state-of-the-art models becomes increasingly high. By synthesizing datasets with high information density, dataset distillation offers a range of potential applications, including support for continual learning, neural architecture search, and privacy protection. Despite recent advances, we lack a holistic understanding of the approaches and applications. Our survey aims to bridge this gap by first proposing a taxonomy of dataset distillation, characterizing existing approaches, and then systematically reviewing the data modalities, and related applications. In addition, we summarize the challenges and discuss future directions for this field of research.

cs.LG↗

arXiv2023

A Comprehensive Study on Dataset Distillation: Performance, Privacy, Robustness and Fairness

The aim of dataset distillation is to encode the rich features of an original dataset into a tiny dataset. It is a promising approach to accelerate neural network training and related studies. Different approaches have been proposed to improve the informativeness and generalization performance of distilled images. However, no work has comprehensively analyzed this technique from a security perspective and there is a lack of systematic understanding of potential risks. In this work, we conduct extensive experiments to evaluate current state-of-the-art dataset distillation methods. We successfully use membership inference attacks to show that privacy risks still remain. Our work also demonstrates that dataset distillation can cause varying degrees of impact on model robustness and amplify model unfairness across classes when making predictions. This work offers a large-scale benchmarking framework for dataset distillation evaluation.

cs.LG↗

arXiv2022

Towards General Deep Leakage in Federated Learning

Unlike traditional central training, federated learning (FL) improves the performance of the global model by sharing and aggregating local models rather than local data to protect the users' privacy. Although this training approach appears secure, some research has demonstrated that an attacker can still recover private data based on the shared gradient information. This on-the-fly reconstruction attack deserves to be studied in depth because it can occur at any stage of training, whether at the beginning or at the end of model training; no relevant dataset is required and no additional models need to be trained. We break through some unrealistic assumptions and limitations to apply this reconstruction attack in a broader range of scenarios. We propose methods that can reconstruct the training data from shared gradients or weights, corresponding to the FedSGD and FedAvg usage scenarios, respectively. We propose a zero-shot approach to restore labels even if there are duplicate labels in the batch. We study the relationship between the label and image restoration. We find that image restoration fails even if there is only one incorrectly inferred label in the batch; we also find that when batch images have the same label, the corresponding image is restored as a fusion of that class of images. Our approaches are evaluated on classic image benchmarks, including CIFAR-10 and ImageNet. The batch size, image quality, and the adaptability of the label distribution of our approach exceed those of GradInversion, the state-of-the-art.

cs.LG↗

arXiv2021

DID-eFed: Facilitating Federated Learning as a Service with Decentralized Identities

We have entered the era of big data, and it is considered to be the "fuel" for the flourishing of artificial intelligence applications. The enactment of the EU General Data Protection Regulation (GDPR) raises concerns about individuals' privacy in big data. Federated learning (FL) emerges as a functional solution that can help build high-performance models shared among multiple parties while still complying with user privacy and data confidentiality requirements. Although FL has been intensively studied and used in real applications, there is still limited research related to its prospects and applications as a FLaaS (Federated Learning as a Service) to interested 3rd parties. In this paper, we present a FLaaS system: DID-eFed, where FL is facilitated by decentralized identities (DID) and a smart contract. DID enables a more flexible and credible decentralized access management in our system, while the smart contract offers a frictionless and less error-prone process. We describe particularly the scenario where our DID-eFed enables the FLaaS among hospitals and research institutions.

cs.CR↗

arXiv2019

Improving Unsupervised Word-by-Word Translation with Language Model and Denoising Autoencoder

Unsupervised learning of cross-lingual word embedding offers elegant matching of words across languages, but has fundamental limitations in translating sentences. In this paper, we propose simple yet effective methods to improve word-by-word translation of cross-lingual embeddings, using only monolingual corpora but without any back-translation. We integrate a language model for context-aware search, and use a novel denoising autoencoder to handle reordering. Our system surpasses state-of-the-art unsupervised neural translation systems without costly iterative training. We also analyze the effect of vocabulary size and denoising type on the translation performance, which provides better understanding of learning the cross-lingual word embedding and its usage in translation.

cs.CL↗