Papers by Haotian Sun

12 papers
Improve Vision Language Model Chain-of-thought Reasoning (2025.acl-long)

Copied to clipboard

Challenge: Current training recipes often rely on datasets dominated by short annotations with limited rationales, hindering the models' ability to generalize to tasks requiring comprehensive reasoning.
Approach: They propose a two-stage post-training strategy that augments short answers with CoT reasoning generated by GPT-4o, enhancing the VLM's CoT capabilities through fine-tuning.
Outcome: The proposed strategy enhances the model's CoT capabilities through fine-tuning and reinforcement learning.
EPO: Hierarchical LLM Agents with Environment Preference Optimization (2024.emnlp-main)

Copied to clipboard

Challenge: Long-horizon decision-making tasks require extensive planning over multiple steps, maintaining coherence and goal orientation, which is difficult for LLMs that are typically designed for more immediate and localized predictions.
Approach: They propose a hierarchical framework that decomposes complex tasks into manageable subgoals, utilizing separate LLMs for subgoal prediction and low-level action generation.
Outcome: The proposed framework achieves first place on the ALFRED public leaderboard and demonstrates its potential to improve long-horizon decision-making in diverse environments.
MedAdapter: Efficient Test-Time Adaptation of Large Language Models Towards Medical Reasoning (2024.emnlp-main)

Copied to clipboard

Challenge: Large language models (LLMs) have improved generation and reasoning capabilities compared to traditional BERT-sized models due to massive number of parameters and extensive pre-training on vast textual corpora.
Approach: They propose a unified post-hoc adapter for test-time adaptation of large language models . they propose to fine-tune only a small BERT-sized adapter to rank candidate LLMs .
Outcome: The proposed adapter improves performance on four biomedical tasks without requiring computational resources or sharing data with third parties.
FACT-E: Causality-Inspired Evaluation for Trustworthy Chain-of-Thought Reasoning (2026.findings-acl)

Copied to clipboard

Challenge: Existing models generate explanations that appear coherent while containing unfaithful intermediate steps.
Approach: They propose a causality-inspired framework for evaluating CoT quality using controlled perturbations as an instrumental signal to separate genuine step-to-step dependence from bias-driven artifacts.
Outcome: Experiments on GSM8K, MATH, and CommonsenseQA show that FACT-E improves reasoning-trajectory selection and yields stronger in-context learning exemplars.
AgentCPM-GUI: Building Mobile-Use Agents with Reinforcement Fine-Tuning (2025.emnlp-demos)

Copied to clipboard

Challenge: Large language model agents have enabled GUI-based automation, but their deployment is limited by noisy data, poor generalization, and lack of support for non-English GUIs.
Approach: They propose an 8B-parameter GUI agent built for robust and efficient on-device GUI interaction.
Outcome: The proposed GUI agent achieves promising performance on five public benchmarks and proposed Chinese benchmark CAGUI.
D-Artemis: A Deliberative Cognitive Framework for Mobile GUI Multi-Agents (2026.findings-acl)

Copied to clipboard

Challenge: Graphical User Interface (GUI) agents aim to automate a wide spectrum of human tasks by emulating user interaction.
Approach: They propose a deliberative framework that leverages a fine-grained tip retrieval mechanism to inform its decision-making process.
Outcome: The proposed framework achieves SOTA among open-source general models on AndroidWorld and ScreenSpot-V2 . it leverages a fine-grained, app-specific tip retrieval mechanism to inform its decision-making process .
Shuttle Between Symbolic Instructions and Neural Parameters of Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Despite their distinct external representations, a deeper analysis reveals their intrinsic nature: instructions serve as a natural language compression devised by humans for data governing specific mapping patterns, whereas parameters act as 'neuro compression' of the same task data.
Approach: They propose a neural network framework to model and learn the bi-directional mappings between instructions and parameters of large language models by evaluating it on the tasks of instruction deduction and induction.
Outcome: The proposed framework can map one of the instructions/parameters to the other by evaluating it on the tasks of instruction deduction and induction.
ItD: Large Language Models Can Teach Themselves Induction through Deduction (2024.acl-long)

Copied to clipboard

Challenge: Recent studies have shown that Large Language Models (LLMs) have limited ability to conduct induction.
Approach: They propose a framework to enable LLMs to teach themselves induction through deduction.
Outcome: The proposed framework improves performance on two induction benchmarks and shows that it can be used to teach induction through deduction.
Grammar-Based Code Representation: Is It a Worthy Pursuit for LLMs? (2025.findings-acl)

Copied to clipboard

Challenge: Existing research demonstrates the effectiveness of grammar-based code representations in small-scale models, showing their ability to reduce syntax errors and enhance performance.
Approach: They develop a series of billion-scale grammar-based code representations that incorporate grammar rules into the code generation process.
Outcome: Experiments on HumanEval and MBPP show that grammar-based representations reduce syntax errors and improve performance even in billion-scale models.
Aligning Large Multimodal Models with Factually Augmented RLHF (2024.findings-acl)

Copied to clipboard

Challenge: Large Multimodal Models (LMMs) are built across modalities and the misalignment between two modality can result in "hallucination" . developing LMMs faces challenges such as a lack of data and a limited number of data sets.
Approach: They propose a new algorithm that augments the reward model with additional factual information such as image captions and ground-truth multi-choice options.
Outcome: The proposed approach improves on the LLaVA-Bench dataset with the 96% performance level of the text-only GPT-4 and an improvement of 60% on MMHAL-BENCH over other baselines.
DebugBench: Evaluating Debugging Capability of Large Language Models (2024.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have demonstrated exceptional coding capabilities, but their debugging capabilities remain relatively unexplored.
Approach: They propose a debugging benchmark consisting of 4,253 LLMs with four major bug categories and 18 minor types in C++, Java, and Python.
Outcome: The proposed benchmark covers four major bug categories and 18 minor types in C++, Java, and Python.
CORES: Code-Oriented Reasoning for Complex Text-to-SQL and Generalizable TableQA (2026.findings-acl)

Copied to clipboard

Challenge: Text-to-SQL models struggle with complex analytical tasks such as generating simple SQL queries.
Approach: They propose a text-to-sql model that leverages Python as a procedural reasoning pivot to enhance both complex SQL generation and tabular reasoning.
Outcome: The proposed model outperforms baseline models on six text-to-SQL benchmarks by 6.44% on average while maintaining good capability on three tableQA benchmarks.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations