17 citations · 20 across the 23 of their papers we have counts for
28 papers
How Do Agentic LLMs Decide to Call Tools? A Tool-Call Vector Shaped by Suppression
Xijie Gong, Tingxu Han, Jiahao Zhang +5
Tool calling, invoking external tools on demand, is central to agentic LLMs, yet the mechanism that decides whether a model calls a tool or responds directly remains poorly underst…
D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders
Xinyue Xu, Jiahao Zhang, Lijie Hu +2
Sparse autoencoders (SAEs) reveal visual structure in diffusion transformers (DiTs), but interpreting a feature does not establish whether it can be used to control generation. We…
RolloutFaith: Auditing Persistent Internal Interventions in Visual World Model
Junchi Yao, Ziyi Wang, Youling Huang +1
Interpretability methods such as probes, activation patches and learned editors are designed to reveal or modify a model's current computation. World models pose a harder requireme…
The Answer-Basin Representation Hypothesis: We Are Not Probing or Steering Concepts
Manjiang Yu, Hongji Li, Zihan Wang +5
Linear probing and activation steering use linear directions to predict and control behavior-level concepts such as correctness, safety, and social bias in question answering. We c…
SharedSAE: One Feature Dictionary Across Language Models
Daniil Ognev, Célian Vasson, Lijie Hu +2
Sparse autoencoders (SAEs) are widely used to interpret language model activations, but SAE training and latent labelling are typically repeated for every model. Here, we show that…
CleanVideo: Adaptive Concept Erasure for Text-to-Video Diffusion Models
Junchi Liao, Hongji Li, Wenrui Zhou +1
Concept erasure aims to selectively eliminate undesired visual semantics from pre-trained generative models without compromising their general utility. Extending concept erasure fr…
Learning Label-Efficient Interpretable Medical Image Diagnosis via Semi-supervised Hypergraph Concept Bottleneck Model
Yijun Yang, Ruiqiang Xiao, Lijie Hu +4
Deep learning has revolutionized medical image analysis, delivering exceptional diagnostic accuracy across diverse applications. Yet, the lack of interpretability in its decision-m…
Matryoshka Concept Bottleneck Models
Ziye Chen, Hongbin Lin, Jie Li +1
Concept Bottleneck Models (CBMs) have emerged as a prominent paradigm for interpretable deep learning, learning by grounding predictions in human-understandable concepts. However,…
Multi-Adapter Representation Interventions via Energy Calibration
Manjiang Yu, Hongji Li, Junwei Chen +4
Representation intervention has emerged as a promising paradigm for aligning large language models toward desired behaviors without modifying model weights. Existing methods typica…
Understanding Generalization and Forgetting in In-Context Continual Learning
Guangyu Li, Meng Ding, Lijie Hu
In-context learning (ICL) derives its power from enabling Large Language Models to adapt to new tasks via prompt-based reasoning alone, entirely bypassing the need for parameter up…
Bayesian Gated Non-Negative Contrastive Learning
Peng Cui, Jiahao Zhang, Lijie Hu
While Contrastive Learning (CL) has revolutionized self-supervised representation learning, its latent representations remain highly entangled and opaque, limiting their interpreta…
MedFM-Robust: Benchmarking Robustness of Medical Foundation Models
Xiangxiang Cui, Tianjin Huang, Yifang Wang +2
Medical foundation models have achieved remarkable clinical performance, yet their robustness under real-world perturbations remains underexplored. We present a robustness benchmar…
From Instance Selection to Fixed-Pool Data Recipe Search for Supervised Fine-Tuning
Haodong Wu, Jiahao Zhang, Lijie Hu +1
Supervised fine-tuning (SFT) data selection is commonly formulated as instance ranking: score each example and retain a top- subset. However, effective SFT training subsets are…
In-Context Learning Operates as Concept Subspace Learning
Wei Tang, Xinyan Jiang, Fakhri Karray +1
Regression and Bayesian accounts of in-context learning (ICL) explain how demonstrations can induce predictors, while mechanistic analyses often identify compact activation directi…
When the Strongest Teacher Is Not the Best Teacher: Student-Centric Answer Selection
Zhengyu Hu, Zheyuan Xiao, Linxin Song +10
LLM training increasingly relies on teacher-generated supervision, from synthetic responses to reasoning traces and tool-use demonstrations. Current practice often chooses the high…
Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images
Qishun Yang, Shu Yang, Lijie Hu +1
Multimodal large language models (MLLMs) face safety misalignment, where visual inputs enable harmful outputs. To address this, existing methods require explicit safety labels or c…
UGID: Unified Graph Isomorphism for Debiasing Large Language Models
Zikang Ding, Junchi Yao, Junhao Li +4
Large language models (LLMs) exhibit pronounced social biases. Output-level or data-optimization--based debiasing methods cannot fully resolve these biases, and many prior works ha…
Functional Subspace Watermarking for Large Language Models
Zikang Ding, Junhao Li, Suling Wu +3
Model watermarking utilizes internal representations to protect the ownership of large language models (LLMs). However, these features inevitably undergo complex distortions during…
Word Recovery in Large Language Models Enables Character-Level Tokenization Robustness
Zhipeng Yang, Shu Yang, Lijie Hu +1
Large language models (LLMs) trained with canonical tokenization exhibit surprising robustness to non-canonical inputs such as character-level tokenization, yet the mechanisms unde…
Less is Enough: Synthesizing Diverse Data in LLM Feature Space with Sparse Autoencoders
Zhongzhi Li, Xuansheng Wu, Yijiang Li +2
The diversity of post-training data is critical for effective downstream performance in large language models (LLMs). Many existing approaches to constructing post-training data qu…
Backdooring CLIP through Concept Confusion
Lijie Hu, Junchi Liao, Weimin Lyu +5
Backdoor attacks pose a serious threat to deep learning models by allowing adversaries to implant hidden behaviors that remain dormant on clean inputs but are maliciously triggered…
Beyond External Monitors: Enhancing Transparency of Large Language Models for Easier Monitoring
Guanxu Chen, Jing Shao, Tao Luo +3
Large language models (LLMs) are becoming increasingly capable, but the mechanisms of their thinking and decision-making processes remain unclear. Chain-of-thoughts (CoTs) have bee…
Backdooring Vision-Language Models with Out-Of-Distribution Data
Weimin Lyu, Jiachen Yao, Saumya Gupta +6
The emergence of Vision-Language Models (VLMs) represents a significant advancement in integrating computer vision with Large Language Models (LLMs) to generate detailed text descr…
A Hopfieldian View-based Interpretation for Chain-of-Thought Reasoning
Lijie Hu, Liang Liu, Shu Yang +6
Chain-of-Thought (CoT) holds a significant place in augmenting the reasoning performance for large language models (LLMs). While some studies focus on improving CoT accuracy throug…