Papers by Chak Tou Leong
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing (2025.acl-long)
Copied to clipboard
Kaishuai Xu, Tiezheng Yu, Wenjun Hou, Yi Cheng, Chak Tou Leong, Liangyou Li, Xin Jiang, Lifeng Shang, Qun Liu, Wenjie Li
| Challenge: | Existing studies to improve mathematical ability typically involve applying preference learning to step-wise solution pairs, but they overlook critical subtle errors. |
| Approach: | They propose a preference learning framework that injects predefined subtle errors into pivotal tokens to construct hard pairs for error mitigation. |
| Outcome: | Extensive experiments show that the proposed framework improves on Qwen2-7B-Instruct and MATH with 4.5K training samples. |
Instruct Once, Chat Consistently in Multiple Rounds: An Efficient Tuning Framework for Dialogue (2024.acl-long)
Copied to clipboard
| Challenge: | Tuning language models for dialogue generation has been a prevalent paradigm for building capable dialogue agents. |
| Approach: | They propose a multi-round interactive dialogue tuning framework that models the speaker roles of agent and user separately. |
| Outcome: | The proposed framework performs superior to fine-tuning and improves dialogue consistency. |
Expanding before Inferring: Enhancing Factuality in Large Language Models through Premature Layers Interpolation (2025.emnlp-main)
Copied to clipboard
Dingwei Chen, Ziqiang Liu, Feiteng Fang, Chak Tou Leong, Shiwen Ni, Ahmadreza Argha, Hamid Alinejad-Rokny, Min Yang, Chengming Li
| Challenge: | Existing approaches to generating factually inconsistent outputs are resource-intensive. |
| Approach: | They propose a plug-and-play intervention designed to enhance factuality by inserting premature layers formed through mathematical interpolation with adjacent layers. |
| Outcome: | The proposed intervention reduces hallucinations while outperforming baselines on four datasets. |
Finding RELIEF: Shaping Reasoning Behavior without Reasoning Supervision via Belief Engineering (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for shaping large reasoning models rely on reinforcement learning or fine-tuning with gold-standard reasoning traces. Existing techniques for behavior shaping rely only on additional reward modeling. |
| Approach: | They propose a framework that aligns a model's self-concept with a target belief blueprint and internalizes desired traits by fine-tuning on synthesized, self-reflective QA pairs that affirm the target belief. |
| Outcome: | The proposed framework outperforms behavior-supervised and preference-based models while requiring significantly lower training costs. |
STeCa: Step-level Trajectory Calibration for LLM Agent Learning (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing work focuses on behavior cloning from expert demonstrations or preference learning through exploratory trajectory sampling, but these methods often struggle to address long-horizon tasks where suboptimal actions accumulate step by step, causing agents to deviate from correct task trajectories. |
| Approach: | They propose a framework for LLM-based agent learning that identifies suboptimal actions through a step-level reward comparison during exploration and constructs calibrated trajectories using LLM reflection. |
| Outcome: | The proposed framework outperforms existing methods in long-horizon tasks where suboptimal actions accumulate step by step, causing agents to deviate from correct task trajectories. |
Muffin: Mitigating Unhelpfulness in Emotional Support Conversations with Multifaceted AI Feedback (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that emotional support conversation models generate unhelpful responses that can hinder their effectiveness. |
| Approach: | They propose a model-agnostic framework called Mitigating unhelpfulness with multifaceted AI feedback for emot io nal support (Muffin) it uses a multifaceted feedback module to assess helpfulness model responses across various facets of emotional support and contrasts helpful and unhelpful responses generated by the model. |
| Outcome: | The proposed framework reduces the likelihood of unhelpful responses by comparing helpful and unhelpfully responses generated by previous models to improve response fluency and relevance. |
KNN-SSD: Enabling Dynamic Self-Speculative Decoding via Nearest Neighbor Layer Set Optimization (2026.findings-eacl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have proven highly capable in handling downstream tasks, but the token-by-token generation in autoregressive decoding results in quadratic computational complexity. |
| Approach: | They propose a method that proposes skipping certain layers to construct a draft model, which eliminates the need for additional parameters or training. |
| Outcome: | The proposed method achieves 1.31.6 speedup in LLM inference while being sensitive to domain shifts. |
TokenSkip: Controllable Chain-of-Thought Compression in LLMs (2025.emnlp-main)
Copied to clipboard
| Challenge: | Chain-of-Thought (CoT) has been proven effective in enhancing the reasoning capabilities of large language models (LLMs). |
| Approach: | They propose a chain-of-thought (CoT) prompting approach that enables LLMs to selectively skip less important tokens, allowing for controllable CoT compression. |
| Outcome: | Experiments show that TokenSkip reduces CoT token usage while preserving strong reasoning performance. |
Seeing Isn’t Believing: Mitigating Belief Inertia via Active Intervention in Embodied Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing large language models (LLMs) have enabled agents to tackle complex embodied tasks through environmental interaction, but they still make suboptimal decisions and perform ineffective actions. |
| Approach: | They propose an active belief intervention mechanism that generates explicit belief states . they characterize belief inertia as a key failure mode of LLM-based agents . |
| Outcome: | The proposed method achieves significant gains in task success rates across embodied benchmarks. |
Merlin’s Whisper: Enabling Efficient Reasoning in Large Language Models via Black-box Persuasive Prompting (2026.acl-long)
Copied to clipboard
| Challenge: | Large reasoning models (LRMs) have demonstrated proficiency in tackling complex tasks through step-by-step thinking. |
| Approach: | They propose a black-box persuasive prompting framework that generates concise responses without compromising accuracy. |
| Outcome: | The proposed framework reduces token usage while preserving performance. |
Why Safeguarded Ships Run Aground? Aligned Large Language Models’ Safety Mechanisms Tend to Be Anchored in The Template Region (2025.acl-long)
Copied to clipboard
| Challenge: | Infilling a fixed template between the input instruction and initial model output is a common practice for existing LLMs, but it is vulnerable to inference-time jailbreak attacks. |
| Approach: | They propose to fill a fixed template between the input instruction and initial model output and to detach safety mechanisms from the template region to mitigate the risk of inference-time jailbreak attacks. |
| Outcome: | The proposed method is widespread across aligned LLMs and shows that it mitigates inference-time jailbreak vulnerabilities. |
Deeper Insights Without Updates: The Power of In-Context Learning Over Fine-Tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Fine-tuning and in-context learning are two prevalent methods in imbuing large language models with task-specific knowledge. |
| Approach: | They propose to use a circuit shift theory to explain why in-context learning is superior to fine-tuning for tasks with implicit patterns. |
| Outcome: | The proposed method can grasp deep patterns and significantly improve accuracy on implicit patterns, compared with fine-tuning and in-context learning. |
E2CL: Exploration-based Error Correction Learning for Embodied Agents (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Language models are exhibiting increasing capability in knowledge utilization and reasoning, but they often suffer from misalignment between their intrinsic knowledge and environmental knowledge, leading to infeasible actions. |
| Approach: | They propose a framework that leverages exploration-induced errors and environmental feedback to enhance environment alignment for embodied agents. |
| Outcome: | The proposed framework outperforms baseline methods and exhibits superior self-correction capabilities. |