arXiv subjects

Xiuyuan Qi

Publications and source records attributed to Xiuyuan Qi.

5 recordsLinked to original sources

arXiv2026

Safe at One Loop, Risky at Another: Aligning Safety Across Recurrent Depths in Looped Language Models

Looped Language Models (LoopLMs) provide a parameter efficient approach to scaling model capabilities through repeated use of shared parameters across recurrent steps. Since each recurrent depth can be read out independently, a single LoopLM exposes a broader output space across inference depths, raising an important question: whether safety is preserved throughout recurrent computation. Prior evaluations suggest that deeper recurrence can improve safety on harmful queries, but robustness under jailbreak attacks remains unclear. We therefore conduct a comprehensive safety evaluation of LoopLMs under jailbreak attacks targeting different recurrent depths. We find that attack success can increase at deeper inference depths, the same query can elicit different safety behaviors across depths, and attacks constructed against one depth can transfer to others. Moreover, SFT and preference alignment do not eliminate these safety gaps, motivating an alignment method designed for LoopLMs. We introduce SafeBridge, which combines lightweight depth specific control of shared recurrent layers, selective state bridging, and joint safety supervision across recurrent depths. Across model scales, multiple attack methods, and safety benchmarks, SafeBridge substantially reduces attack success for both matched-depth and cross-depth attacks. It also improves general utility over the vanilla models while maintaining comparable over-refusal behavior. Our results show that the safety of a LoopLM cannot be inferred from a single recurrent depth, motivating safety alignment across recurrent computation. Our code and model checkpoints will be released upon acceptance.

cs.CR↗

arXiv2026

Differential Privacy Protected Preference Alignment on Survey Derived Data

Preference alignment enables language models to better reflect human values, but the preference data contains sensitive information, especially when derived from surveys about political beliefs, social norms, religion, gender attitudes, and cultural values. This paper studies privacy-preserving alignment on questionnaire-derived data from the World Values Survey, where supervision takes the form of ordinal labels rather than standard pairwise rankings. Existing efforts mainly focus on general privacy protections upon model training, while data-specific methods for survey data remain unexplored. We first show the vulnerability of aligned models, then formulate the task as rating prediction and introduce label differential privacy. Building on this formulation, we construct a differential private pipeline utilizing three privacy mechanisms: randomized response, exponential mechanism, and DP-SGD. Experiments show that simple label-level mechanisms outperform gradient-level methods, and proper aggregation function for preprocessing also make large contribution. The new fine-grained private comparison motivates explorations toward language model alignment with formal privacy guarantees.

cs.CR↗

arXiv2026

Beyond Reproducibility: Token Probabilities Expose Large Language Model Nondeterminism

The execution of Large Language Models (LLMs) has been shown to produce nondeterministic results when run on Graphics Processing Units (GPUs), even when they are configured to produce deterministic results. This is due to the finite precision effects of the arithmetic operations, which depend on the order in which they are executed. This order, in turn, depends on the processes that are running concurrently on the GPU. Previous studies have focused on the impact of nondeterminism on the text generated by the LLMs or on proposing mechanisms to achieve deterministic execution. This work takes a closer look at nondeterminism by analyzing the variations on the token probabilities, not on the generated text. Interestingly, all the models evaluated have similar results in both the trends and the actual values of the variations of the probabilities. In particular, the results show that the effects of nondeterminism are significant for token probabilities that are in the range of 0.1 to 0.9, while they are much smaller when the probabilities are close to 0 or 1. This has significant implications for our understanding of nondeterminism. The first is that nondeterminism will likely have a non-negligible impact on generated text when the temperature is not zero, as it introduces significant variations in the token probabilities except when they are close to 0 or 1. Secondly, it suggests that all models have similar non deterministic variations at the token probability level. Therefore, different variations in the performance of the generated text, for example, when measuring accuracy on a benchmark, seem to come from different token probabilities or response lengths. A third implication is that we may be able to estimate the impact of nondeterminism by running a single inference and analyzing the token level probabilities, instead of having to run the same inference many times.

cs.AI↗

arXiv2024

Cross-modality Information Check for Detecting Jailbreaking in Multimodal Large Language Models

Multimodal Large Language Models (MLLMs) extend the capacity of LLMs to understand multimodal information comprehensively, achieving remarkable performance in many vision-centric tasks. Despite that, recent studies have shown that these models are susceptible to jailbreak attacks, which refer to an exploitative technique where malicious users can break the safety alignment of the target model and generate misleading and harmful answers. This potential threat is caused by both the inherent vulnerabilities of LLM and the larger attack scope introduced by vision input. To enhance the security of MLLMs against jailbreak attacks, researchers have developed various defense techniques. However, these methods either require modifications to the model's internal structure or demand significant computational resources during the inference phase. Multimodal information is a double-edged sword. While it increases the risk of attacks, it also provides additional data that can enhance safeguards. Inspired by this, we propose Cross-modality Information DEtectoR (CIDER), a plug-and-play jailbreaking detector designed to identify maliciously perturbed image inputs, utilizing the cross-modal similarity between harmful queries and adversarial images. CIDER is independent of the target MLLMs and requires less computation cost. Extensive experimental results demonstrate the effectiveness and efficiency of CIDER, as well as its transferability to both white-box and black-box MLLMs.

cs.CL↗

arXiv2024

Performance evaluation of Reddit Comments using Machine Learning and Natural Language Processing methods in Sentiment Analysis

Sentiment analysis, an increasingly vital field in both academia and industry, plays a pivotal role in machine learning applications, particularly on social media platforms like Reddit. However, the efficacy of sentiment analysis models is hindered by the lack of expansive and fine-grained emotion datasets. To address this gap, our study leverages the GoEmotions dataset, comprising a diverse range of emotions, to evaluate sentiment analysis methods across a substantial corpus of 58,000 comments. Distinguished from prior studies by the Google team, which limited their analysis to only two models, our research expands the scope by evaluating a diverse array of models. We investigate the performance of traditional classifiers such as Naive Bayes and Support Vector Machines (SVM), as well as state-of-the-art transformer-based models including BERT, RoBERTa, and GPT. Furthermore, our evaluation criteria extend beyond accuracy to encompass nuanced assessments, including hierarchical classification based on varying levels of granularity in emotion categorization. Additionally, considerations such as computational efficiency are incorporated to provide a comprehensive evaluation framework. Our findings reveal that the RoBERTa model consistently outperforms the baseline models, demonstrating superior accuracy in fine-grained sentiment classification tasks. This underscores the substantial potential and significance of the RoBERTa model in advancing sentiment analysis capabilities.

cs.CL↗