1 citations · 1 across the 5 of their papers we have counts for
7 papers
Rethinking Reasoning Paths as Phase-Structured Trajectories
Zhenghao He, Guangzhi Xiong, Sanchit Sinha +3
Large language models often improve problem-solving performance by generating multi-step reasoning paths, yet how to analyze the hidden states along these paths remains unclear. Ex…
Reliable Parallel Decoding in Masked Diffusion Language Models
Zhenghao He, Bohan Liu, Guangzhi Xiong +1
Masked diffusion language models (MDLMs) can generate text efficiently by predicting multiple masked tokens in parallel, but predictions from the same forward pass are not necessar…
Towards Robustness against Typographic Attack with Training-free Concept Localization
Bohan Liu, Wenqian Ye, Guangzhi Xiong +3
Models trained via Contrastive Language-Image Pretraining (CLIP) serve as the foundational vision encoders for most modern Large Vision Language Models (LVLMs). Despite their wides…
Attention-guided Fine-tuning of Multimodal Large Language Models Improves Chain-of-Thought Reasoning
Sanchit Sinha, Guangzhi Xiong, Bohan Liu +2
The effectiveness of Chain-of-Thought (CoT) prompting in Multimodal Large Language Models (MLLMs) remains uncertain: across several visual reasoning benchmarks, CoT prompting often…
Triggering Chain-of-Thought via Latent Feature Interventions in Large Language Models
Zhenghao He, Guangzhi Xiong, Bohan Liu +2
Chain-of-Thought (CoT) prompting often improves the reasoning performance of large language models (LLMs), but the internal signal that triggers this behavior remains poorly unders…
Toward Faithful Retrieval-Augmented Generation with Sparse Autoencoders
Guangzhi Xiong, Zhenghao He, Bohan Liu +2
Retrieval-Augmented Generation (RAG) improves the factuality of large language models (LLMs) by grounding outputs in retrieved evidence, but faithfulness failures, where generation…
SAGE: Spuriousness-Aware Guided Prompt Exploration for Mitigating Multimodal Bias
Wenqian Ye, Di Wang, Guangtao Zheng +2
Large vision-language models, such as CLIP, have shown strong zero-shot classification performance by aligning images and text in a shared embedding space. However, CLIP models oft…