Researcher — paper.dinq.me

Sparse Autoencoders (SAEs) are a prominent tool in mechanistic interpretability (MI) for decomposing neural network activations into interpretable features. However, the aspiration to identify a canonical set of features is challenged by the observed inconsistency of learned SAE features across d...

可解释性效率与系统评测与基准OpenReview 项目页

作者位次 #7/7ACL 2026Poster· Natural Language Processing

Advancing Reasoning in Diffusion Language Models with Denoising Process Rewards

Shaoan Xie · Lingjing Kong · Xiangchen Song · Xinshuai Dong · Guangyi Chen · Eric P. Xing · Kun Zhang

摘要

Diffusion-based large language models offer a non-autoregressive alternative for text generation, but enabling them to perform complex reasoning remains challenging. Reinforcement learning has recently emerged as an effective post-training strategy for improving their performance; however, existi...

强化学习与后训练推理可解释性OpenReview 项目页

作者位次 #8/8ACL 2026Findings· Findings of ACL

Text Embedding as Treatment: A Meta Causal Approach for Robust Sentiment Classification

Fengxiang Cheng · Chuan Zhou · Xiang Li · Haoxuan Li · Wen-li Wang · Jinkun Chen · Mingming Gong · Kun Zhang

摘要

Sentiment classification is a crucial task in natural language processing (NLP). To mitigate the spurious correlation, the causal word identification method estimates the impact of treatment words on sentence sentiment and removes those with low treatment effects. However, previous works regard t...

数据与预训练可解释性理论与学习动力学OpenReview 项目页

作者位次 #4/6AAAI 2026Poster· Artificial Intelligence

Socrates or Smartypants: Testing Logic Reasoning Capabilities of Large Language Models with Logic Programming-Based Test Oracles

Zihao Xu · Junchen Ding · Yiling Lou · Kun Zhang · Dong Gong · Yuekang Li

推理代码OpenReview

Kun Zhang

获取完整研究档案

联系方式

2026 顶会论文

Distributional Equivalence in Linear Non-Gaussian Latent-Variable Cyclic Causal Models: Characterization and Learning

Independence Test for Linear Non-Gaussian Data and Applications in Causal Discovery

Conditional Independent Component Analysis for Estimating Causal Structure with Latent Variables

On the Thinking-Language Modeling Gap in Large Language Models

Characterization and Learning of Causal Graphs with Latent Confounders and Post-treatment Selection from Interventional Data

Selection, Reflection and Self-Refinement: Revisit Reasoning Tasks via a Causal Lens

Diverse Dictionary Learning

Ada-Diffuser: Latent-Aware Adaptive Diffusion for Decision-Making

Beyond DAGs: A Latent Partial Causal Model for Multimodal Learning

Controllable Video Generation with Provable Disentanglement

PersonaX: Multimodal Datasets with LLM-Inferred Behavior Traits

Score-based Greedy Search for Structure Identification of Partially Observed Causal Models

Learning by Analogy: A Causal Framework for Compositional Generalization

Mechanistic Interpretability Should Prioritize Feature Consistency in Sparse Autoencoders

Advancing Reasoning in Diffusion Language Models with Denoising Process Rewards

Text Embedding as Treatment: A Meta Causal Approach for Robust Sentiment Classification

Socrates or Smartypants: Testing Logic Reasoning Capabilities of Large Language Models with Logic Programming-Based Test Oracles

常见合作者