Research Work
Selected ongoing and recent research work. Some manuscripts are currently under review, so the descriptions below are intentionally concise and will be updated as results become public.
BizCompass: LLM Reasoning in Business Knowledge and Applications
2026Accepted by ACL
- Constructed a benchmark for evaluating the reasoning capabilities of large language models on business knowledge and application tasks.
- Analyzed model behavior across tasks and identified potential directions for improving business-domain reasoning.
LLM EvaluationBenchmarkReasoning
Unified-MAS: Automatic Domain-Specific Node Generation for Multi-Agent Systems
2026ACL ARR 2026 March submission
- Proposed a unified approach for automatically generating domain-specific nodes to empower multi-agent systems.
- Focused on making multi-agent systems more adaptable across tasks and domains.
Multi-Agent SystemsAutomatic GenerationLLM Agents
AMA: Adaptive Memory via Multi-Agent Collaboration
2026Accepted by ACL
- Introduced an adaptive memory framework based on multi-agent collaboration.
- Explored how collaborative agents can improve memory retrieval and utilization in large language models.
Agent MemoryMulti-Agent CollaborationLLMs
TiARA: Time-Accumulated Rank Adaptation for Constant-Cost Fine-Tuning
2026ICML 2026 conference submission
- Developed a time-accumulated rank adaptation technique for constant-cost fine-tuning.
- Aimed to reduce computational requirements while maintaining model performance.
Fine-TuningParameter-Efficient LearningLLMs
Unnoticeable Graph Injection Attack on Graph Neural Networks
2024Accepted by IJCNN
- Investigated stealthy graph injection attacks on graph neural networks.
- Developed UNGIA, a method that selects representative nodes and uses bi-level optimization to balance homogeneity and attack effectiveness.
- Conducted experiments on Cora, Citeseer, and Cora-ML benchmark datasets.
Graph Neural NetworksSecurityAdversarial Learning
CNO-former and Chaos-Based Neural Models
2023 - 2024Selected undergraduate research projects
- Formulated CNO-former, a Transformer-style model that modifies neuron states within self-attention to reduce computational cost and mitigate information loss.
- Explored chaotic neural oscillation mechanisms for aspect-based sentiment analysis.
- Awarded an Excellent Award in the 2nd Computer Science Related Poster Exhibition.
TransformerSelf-AttentionSentiment Analysis