Papers by Wenxiao Wang
Explore the Reasoning Capability of LLMs in the Chess Testbed (2025.naacl-short)
Copied to clipboard
| Challenge: | a recent study shows that large language models struggle with long-term, complex reasoning tasks. |
| Approach: | They propose to integrate annotated strategy and tactic into large language models to improve reasoning capability. |
| Outcome: | The proposed model performs better than GPT, Claude, and Gemini models . it integrates annotated strategy and tactic into the model . |
HeteroCache: A Dynamic Retrieval Approach to Heterogeneous KV Cache Compression for Long-Context LLM Inference (2026.acl-long)
Copied to clipboard
| Challenge: | Existing static compression methods suffer from coarse-grained caching and high I/O overhead. |
| Approach: | They propose a training-free dynamic compression framework that uses a sparse attention mechanism to categorize attention heads based on stability and similarity. |
| Outcome: | The proposed framework achieves state-of-the-art performance on long-context benchmarks and accelerates decoding by up to 3 compared to the original model with a 224K context. |
Tool Preferences in Agentic LLMs are Unreliable (2025.emnlp-main)
Copied to clipboard
Kazem Faghih, Wenxiao Wang, Yize Cheng, Siddhant Bharti, Gaurang Sriramanan, Sriram Balasubramanian, Parsa Hosseini, Soheil Feizi
| Challenge: | Large language models (LLMs) can now access a wide range of external tools thanks to the Model Context Protocol (MCP). |
| Approach: | They expose a vulnerability in prevalent tool/function-calling protocols by editing tool descriptions to find out which tools are used by LLMs. |
| Outcome: | The proposed changes in the tool descriptions can increase the usage of tools from LLMs when competing with alternatives. |
From Redundancy to Relevance: Information Flow in LVLMs Across Reasoning Tasks (2025.naacl-long)
Copied to clipboard
Xiaofeng Zhang, Yihao Quan, Chen Shen, Xiaosong Yuan, Shaotian Yan, Liang Xie, Wenxiao Wang, Chaochen Gu, Hao Tang, Jieping Ye
| Challenge: | Large visionlanguage models (LVLMs) are a powerful visual-language reasoning tool. |
| Approach: | They propose to integrate attention analysis with LLaVA-CAM to determine interactions between visual representations. |
| Outcome: | The proposed approach can be used to determine interactions between visual representations. |
DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors (2025.emnlp-main)
Copied to clipboard
| Challenge: | Open benchmarks are essential for evaluating large language models, but their accessibility makes them likely targets of test set contamination. |
| Approach: | They propose a framework that leverages backdoor attacks to flag models that used benchmark test sets during training. |
| Outcome: | The proposed framework detects models that trained on benchmark test sets without loss of logits or internal details . it can prevent false accusations while providing strong evidence for every detected case of contamination. |
Your LLM Agents are Temporally Blind: The Misalignment Between Tool Use Decisions and Human Time Perception (2026.findings-acl)
Copied to clipboard
Yize Cheng, Arshia Soltani Moakhar, Chenrui Fan, Parsa Hosseini, Kazem Faghih, Zahra Sodagar, Wenxiao Wang, Soheil Feizi
| Challenge: | Large language model agents assume a stationary context, failing to account for real-world time elapsed between messages. |
| Approach: | They construct a dataset of multi-turn user–agent message trajectories across 76 scenarios . they collect human preferences between "calling a tool" and "directly answering" they also examine whether existing models lack human temporal perception . |
| Outcome: | The results show that existing models display poor alignment with human temporal perception . the findings provide insights to foster the development of more time-aware and human-aligned agents. |
Concise and Organized Perception Facilitates Reasoning in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Extensive experimental results on several popular logical benchmarks (ProofWriter, PrOntoQA, PrONtoQA-OOD, and FOLIO) and mathematical benchmark (DI-GSM) show that COP significantly outperforms previous state-of-the-art methods. |
| Approach: | They propose a reasoning approach called Concise and Organized Perception (COP) that carefully analyzes the given statements to identify the most pertinent information while eliminating redundancy efficiently. |
| Outcome: | The proposed approach outperforms state-of-the-art methods on several popular logical benchmarks and mathematical benchmarks. |