arXiv subjects

Shaohan Yu

Publications and source records attributed to Shaohan Yu.

3 recordsLinked to original sources

arXiv2025

ProGuard: Towards Proactive Multimodal Safeguard

The rapid evolution of generative models has led to a continuous emergence of multimodal safety risks, exposing the limitations of existing defense methods. To address these challenges, we propose ProGuard, a vision-language proactive guard that identifies and describes out-of-distribution (OOD) safety risks without the need for model adjustments required by traditional reactive approaches. We first construct a modality-balanced dataset of 87K samples, each annotated with both binary safety labels and risk categories under a hierarchical multimodal safety taxonomy, effectively mitigating modality bias and ensuring consistent moderation across text, image, and text-image inputs. Based on this dataset, we train our vision-language base model purely through reinforcement learning (RL) to achieve efficient and concise reasoning. To approximate proactive safety scenarios in a controlled setting, we further introduce an OOD safety category inference task and augment the RL objective with a synonym-bank-based similarity reward that encourages the model to generate concise descriptions for unseen unsafe categories. Experimental results show that ProGuard achieves performance comparable to closed-source large models on binary safety classification, substantially outperforms existing open-source guard models on unsafe content categorization. Most notably, ProGuard delivers a strong proactive moderation ability, improving OOD risk detection by 52.6% and OOD risk description by 64.8%.

cs.CV↗

arXiv2025

SafeRBench: Dissecting the Reasoning Safety of Large Language Models

Large Reasoning Models (LRMs) have significantly improved problem-solving through explicit Chain-of-Thought (CoT) reasoning. However, this capability creates a Safety-Helpfulness Paradox: the reasoning process itself can be misused to justify harmful actions or conceal malicious intent behind lengthy intermediate steps. Most existing benchmarks only check the final output, missing how risks evolve, or ``drift'', during the model's internal reasoning. To address this, we propose SafeRBench, the first framework to evaluate LRM safety end-to-end, from the initial input to the reasoning trace and final answer. Our approach introduces: (i) a Risk Stratification Probing that uses specific risk levels to stress-test safety boundaries beyond simple topics; (ii) Micro-Thought Analysis, a new chunking method that segments traces to pinpoint exactly where safety alignment breaks down; and (iii) a comprehensive suite of 10 fine-grained metrics that, for the first time, jointly measure a model's Risk Exposure (e.g., risk level, execution feasibility) and Safety Awareness (e.g., intent awareness). Experiments on 19 LRMs reveal that while enabling Thinking modes improves safety in mid-sized models, it paradoxically increases actionable risks in larger models due to a strong always-help tendency.

cs.AI↗

arXiv2024

HiMoE: Heterogeneity-Informed Mixture-of-Experts for Fair Spatial-Temporal Forecasting

Achieving both accurate and consistent predictive performance across spatial nodes is crucial for ensuring the validity and reliability of outcomes in fair spatial-temporal forecasting tasks. However, existing training methods treat heterogeneous nodes with a fully averaged perspective, resulting in inherently biased prediction targets. Balancing accuracy and consistency is particularly challenging due to the multi-objective nature of spatial-temporal forecasting. To address this issue, we propose a novel Heterogeneity-Informed Mixture-of-Experts (HiMoE) framework that delivers both uniform and precise spatial-temporal predictions. From a model architecture perspective, we design the Heterogeneity-Informed Graph Convolutional Network (HiGCN) to address trend heterogeneity, and we introduce the Node-wise Mixture-of-Experts (NMoE) module to handle cardinality heterogeneity across nodes. From an evaluation perspective, we propose STFairBench, a benchmark that handles fairness in spatial-temporal prediction from both training and evaluation stages. Extensive experiments on four real-world datasets demonstrate that HiMoE achieves state-of-the-art performance, outperforming the best baseline by at least 9.22% across all evaluation metrics.

cs.LG↗