Yue Xu

Subrat Kishore Dutta, Yuelin Xu, Piyush Pant et al.

Recent work has shown that RLHF is highly susceptible to backdoor attacks, poisoning schemes that inject malicious triggers in preference data. However, existing methods often rely on static, rare-token-based triggers, limiting their effectiveness in realistic scenarios. In this paper, we develop GREAT, a novel framework for crafting generalizable backdoors in RLHF through emotion-aware trigger synthesis. Specifically, GREAT targets harmful response generation for a vulnerable user subgroup characterized by both semantically violent requests and emotionally angry triggers. At the core of GREAT is a trigger identification pipeline that operates in the latent embedding space, leveraging principal component analysis and clustering techniques to identify the most representative triggers. To enable this, we present Erinyes, a high-quality dataset of over $5000$ angry triggers curated from GPT-4.1 using a principled, hierarchical, and diversity-promoting approach. Experiments on benchmark RLHF datasets demonstrate that GREAT significantly outperforms baseline methods in attack success rates, especially for unseen trigger scenarios, while largely preserving the response quality on benign inputs.

Yue Xu, Wenjie Wang

Multimodal large language models (MLLMs) have shown impressive capabilities across tasks involving both visual and textual modalities. However, growing concerns remain about their potential to encode and amplify gender bias, particularly in socially sensitive applications. Existing benchmarks predominantly evaluate bias in isolated scenarios, overlooking how bias may emerge subtly through interpersonal interactions. We fill this gap by going beyond single-entity evaluation and instead focusing on a deeper examination of relational and contextual gender bias in dual-individual interactions. We introduce Genres, a novel benchmark designed to evaluate gender bias in MLLMs through the lens of social relationships in generated narratives. Genres assesses gender bias through a dual-character profile and narrative generation task that captures rich interpersonal dynamics and supports a fine-grained bias evaluation suite across multiple dimensions. Experiments on both open- and closed-source MLLMs reveal persistent, context-sensitive gender biases that are not evident in single-character settings. Our findings underscore the importance of relationship-aware benchmarks for diagnosing subtle, interaction-driven gender bias in MLLMs and provide actionable insights for future bias mitigation.

Somrita Ghosh, Yuelin Xu, Xiao Zhang

Achieving high model robustness under adversarial settings is widely recognized as demanding considerable training samples. Recent works propose semi-supervised adversarial training (SSAT) methods with external unlabeled or synthetically generated data, which are the current state-of-the-art. However, SSAT requires substantial extra data to attain high robustness, resulting in prolonged training time and increased memory usage. In this paper, we propose unlabeled data reduction strategies to improve the efficiency of SSAT. Specifically, we design novel latent clustering-based techniques to select or generate a small critical subset of data samples near the model's decision boundary. While focusing on boundary-adjacent points, our methods maintain a balanced ratio between boundary and non-boundary data points to avoid overfitting. Comprehensive experiments on benchmark datasets demonstrate that our methods can significantly reduce SSAT's data requirement and computation costs while preserving its strong robustness advantages. In particular, our latent-space selection scheme based on k-means clustering and our guided DDPM fine-tuning approach with LCG-KM are the most effective, achieving nearly identical robust accuracies with 5x to 10x less unlabeled data and approximately 4x less total runtime.