arXiv subjects

Weijuan Zhang

Publications and source records attributed to Weijuan Zhang.

8 recordsLinked to original sources

arXiv2026

JailMeter: An Evidence-Based Evaluation Framework for Jailbreak Attacks on Large Language Models

The assessment of jailbreak attacks against large language models currently suffers from inconsistent evaluation criteria and methods, leading to unreliable estimates of attack success rates. We propose JailMeter, an evidence-based evaluation framework designed to more faithfully measure jailbreak effectiveness. Inspired by the Information Bottleneck theory, JailMeter applies dual-feedback optimization to filter jailbreak noise from model responses while preserving content relevant to the original malicious question. This process produces concise evidence for a rigorous assessment under which an attack is validated only when the response captures the malicious intent and delivers a complete answer, thereby signaling a substantive bypass of model safety alignment. We evaluate JailMeter on JailMeter-Eva, a challenging benchmark containing 330 human-labeled, non-rejected jailbreak instances. JailMeter achieves an accuracy of 97.27%, substantially outperforming existing evaluation methods. To support large-scale evaluation, we further distill JailMeter into a small language model, JailMeter\textsubscript{SLM}, which maintains comparable reliability with significantly reduced computational costs. Code and dataset are available at https://github.com/Magi2B0y/JailMeter.

cs.CR↗

arXiv2025

Focusing on Language: Revealing and Exploiting Language Attention Heads in Multilingual Large Language Models

Large language models (LLMs) increasingly support multilingual understanding and generation. Meanwhile, efforts to interpret their internal mechanisms have emerged, offering insights to enhance multilingual performance. While multi-head self-attention (MHA) has proven critical in many areas, its role in multilingual capabilities remains underexplored. In this work, we study the contribution of MHA in supporting multilingual processing in LLMs. We propose Language Attention Head Importance Scores (LAHIS), an effective and efficient method that identifies attention head importance for multilingual capabilities via a single forward and backward pass through the LLM. Applying LAHIS to Aya-23-8B, Llama-3.2-3B, and Mistral-7B-v0.1, we reveal the existence of both language-specific and language-general heads. Language-specific heads enable cross-lingual attention transfer to guide the model toward target language contexts and mitigate off-target language generation issue, contributing to addressing challenges in multilingual LLMs. We also introduce a lightweight adaptation that learns a soft head mask to modulate attention outputs over language heads, requiring only 20 tunable parameters to improve XQuAD accuracy. Overall, our work enhances both the interpretability and multilingual capabilities of LLMs from the perspective of MHA.

cs.CL↗

arXiv2025

Prefill-level Jailbreak: A Black-Box Risk Analysis of Large Language Models

Large Language Models face security threats from jailbreak attacks. Existing research has predominantly focused on prompt-level attacks while largely ignoring the underexplored attack surface of user-controlled response prefilling. This functionality allows an attacker to dictate the beginning of a model's output, thereby shifting the attack paradigm from persuasion to direct state manipulation.In this paper, we present a systematic black-box security analysis of prefill-level jailbreak attacks. We categorize these new attacks and evaluate their effectiveness across fourteen language models. Our experiments show that prefill-level attacks achieve high success rates, with adaptive methods exceeding 99% on several models. Token-level probability analysis reveals that these attacks work through initial-state manipulation by changing the first-token probability from refusal to compliance.Furthermore, we show that prefill-level jailbreak can act as effective enhancers, increasing the success of existing prompt-level attacks by 10 to 15 percentage points. Our evaluation of several defense strategies indicates that conventional content filters offer limited protection. We find that a detection method focusing on the manipulative relationship between the prompt and the prefill is more effective. Our findings reveal a gap in current LLM safety alignment and highlight the need to address the prefill attack surface in future safety training.

cs.CR↗

arXiv2025

Latent Knowledge Scalpel: Precise and Massive Knowledge Editing for Large Language Models

Large Language Models (LLMs) often retain inaccurate or outdated information from pre-training, leading to incorrect predictions or biased outputs during inference. While existing model editing methods can address this challenge, they struggle with editing large amounts of factual information simultaneously and may compromise the general capabilities of the models. In this paper, our empirical study demonstrates that it is feasible to edit the internal representations of LLMs and replace the entities in a manner similar to editing natural language inputs. Based on this insight, we introduce the Latent Knowledge Scalpel (LKS), an LLM editor that manipulates the latent knowledge of specific entities via a lightweight hypernetwork to enable precise and large-scale editing. Experiments conducted on Llama-2 and Mistral show even with the number of simultaneous edits reaching 10,000, LKS effectively performs knowledge editing while preserving the general abilities of the edited LLMs. Code is available at: https://github.com/Linuxin-xxx/LKS.

cs.LG↗

arXiv2021

Flips on homologous orientations of surface graphs with prescribed forbidden facial circuits

Let $G$ be a graph embedded on an orientable surface. Given a class ${\cal C}$ of facial circuits of $G$ as a forbidden class, we give a sufficient-necessary condition for that an $α$-orientation (orientation with prescribed out-degrees) of $G$ can be transformed into another by a sequence of flips on non-forbidden circuits and further give an explicit formula for the minimum number of such flips. We also consider the connection among all $α$-orientations by defining a directed graph ${\bf D}({\cal C})$, namely the ${\cal C}$-forbidden flip graph. We show that if ${\cal C}\not=\emptyset$, then ${\bf D}({\cal C})$ has exactly $|O(G,{\cal C})|$ components, each of which is the cover graph of a distributive lattice, where $|O(G,{\cal C})|$ is the number of the $α$-orientations that has no counterclockwise facial circuit other than that in ${\cal C}$. If ${\cal C}=\emptyset$, then every component of ${\bf D}({\cal C})$ is strongly connected. This generalizes the corresponding results of Felsner and Propp for the case that ${\cal C}$ consists of a single facial circuit.

math.CO↗

arXiv2020

SEEF-ALDR: A Speaker Embedding Enhancement Framework via Adversarial Learning based Disentangled Representation

Speaker verification, as a biometric authentication mechanism, has been widely used due to the pervasiveness of voice control on smart devices. However, the task of "in-the-wild" speaker verification is still challenging, considering the speech samples may contain lots of identity-unrelated information, e.g., background noise, reverberation, emotion, etc. Previous works focus on optimizing the model to improve verification accuracy, without taking into account the elimination of the impact from the identity-unrelated information. To solve the above problem, we propose SEEF-ALDR, a novel Speaker Embedding Enhancement Framework via Adversarial Learning based Disentangled Representation, to reinforce the performance of existing models on speaker verification. The key idea is to retrieve as much speaker identity information as possible from the original speech, thus minimizing the impact of identity-unrelated information on the speaker verification task by using adversarial learning. Experimental results demonstrate that the proposed framework can significantly improve the performance of speaker verification by 20.3% and 23.8% on average over 13 tested baselines on dataset Voxceleb1 and 8 tested baselines on dataset Voxceleb2 respectively, without adjusting the structure or hyper-parameters of them. Furthermore, the ablation study was conducted to evaluate the contribution of each module in SEEF-ALDR. Finally, porting an existing model into the proposed framework is straightforward and cost-efficient, with very little effort from the model owners due to the modular design of the framework.

eess.AS↗

arXiv2020

ET-GAN: Cross-Language Emotion Transfer Based on Cycle-Consistent Generative Adversarial Networks

Despite the remarkable progress made in synthesizing emotional speech from text, it is still challenging to provide emotion information to existing speech segments. Previous methods mainly rely on parallel data, and few works have studied the generalization ability for one model to transfer emotion information across different languages. To cope with such problems, we propose an emotion transfer system named ET-GAN, for learning language-independent emotion transfer from one emotion to another without parallel training samples. Based on cycle-consistent generative adversarial network, our method ensures the transfer of only emotion information across speeches with simple loss designs. Besides, we introduce an approach for migrating emotion information across different languages by using transfer learning. The experiment results show that our method can efficiently generate high-quality emotional speech for any given emotion category, without aligned speech pairs.

cs.SD↗

arXiv2017

Flip-distance between α-orientations of graphs embedded on plane and sphere

Felsner introduced a cycle reversal, namely the `flip' reversal, for α-orientations (i.e., each vertex admits a prescribed out-degree) of a graph G embedded on the plane and further proved that the set of all the α-orientations of G carries a distributive lattice with respect to the flip reversals. In this paper, we give an explicit formula for the minimum number of flips needed to transform one α-orientation into another for graphs embedded on the plane or sphere, respectively.

math.CO↗