Zhenke Duan

M.S. in Applied Statistics (Data Science)

Scroll
ZD

About Me

I hold a B.S. in Financial Mathematics and an M.S. in Applied Statistics (Data Science) from ZUEL. My research centers on Natural Language Processing, Affective Computing, and cognitive-inspired multi-agent reasoning. I have hands-on experience with open-source projects and am proficient in PyTorch, TensorFlow, LangChain, and vLLM.

"Water benefits all things without contention."

I also have strong Linux-based full-stack development skills and programming experience in Python, R, and Java. I am passionate about building open-source research platforms and applying machine learning to solve real-world problems.

M.S. in Applied Statistics Zhongnan University of Economics and Law, 2024 - Present
B.S. in Financial Mathematics Zhongnan University of Economics and Law, 2020 - 2024
Institute Institute for AI Optimization and Decision Science (AIODS), ZUEL
Founder & CTO CodeSoul.co (Hangzhou Lingmu Chuangxiang Technology Co., Ltd.)

Research Interests

Natural Language Processing

Developing techniques for text understanding, generation, and semantic analysis.

Affective Computing

Research on emotion recognition, sentiment analysis, and human-computer emotional interaction.

Multi-Agent Systems

Cognitive-inspired reasoning and communication optimization for multi-agent collaboration.

Large Language Models

Chain-of-thought reasoning, prompt engineering, and LLM agent frameworks.

Statistical Learning

Advanced statistical methods, multivariate analysis, and time series modeling.

Cognitive Science

LLM cognition, mental health AI, and human-AI interaction.

Publications

2026

CRFormer: Covariance-Enhanced Rough Transformer for Multivariate Time Series Modeling

Submitted to NeurIPS 2026, First Author

Read Paper →
2026

Data-Adaptive Mahalanobis Metric Learning for Cross-Head Attention in Transformers

Submitted to NeurIPS 2026, First Author

Read Paper →
2026

Structural Alignment of Mental Health Categories in LLM Representation Space

Submitted to EMNLP 2026, Corresponding Author

Read Paper →
2026

Tree-CoT-RT: An Explainable Multi-Path Tree-Guided Chain-of-Thought and Reinforcement Learning Framework for Aspect Sentiment Quad Prediction

Findings of ACL 2026, Third Author

Read Paper →
2026

Building a causality-aware single-cell RNA-Seq foundation model via context-specific causal regulation modeling

Submitted to Nature Biomedical Engineering, Fourth Author

Read Paper →
2026

BEAM: A Bayesian Energy-Aware Framework for Multi-Agent Communication Optimization under Incomplete Information

IEEE Transactions on Cybernetics (SCI-Q1, CCF-B), First Author

Read Paper →
2025

ECCoT: A Framework for Enhancing Effective Cognition via Chain of Thought in Large Language Models

Proceedings of the Cognitive Science Society (CogSci) 2025 (Vol. 47), CCF-B, First Author

Read Paper →
2025

KG-MASD: Knowledge-Graph-Guided Multi-Agent Collaboration and Distillation

Information Systems (SCI-Q2, CCF-B), Co-First Author

Paper → GitHub →

Projects

Plasmod: Agent-Native Database for Multi-Agent Systems

GitHub →

Developed an open-source agent-native database system for multi-agent applications, unifying cognitive object storage, event-driven materialization, and structured evidence retrieval within a single runnable Go server.

THETA: LLM-Enhanced Topic Analysis Platform

GitHub →

Developed an open-source research platform for topic analysis in social science, integrating domain-adaptive document embeddings from Qwen-3 models with generative topic modeling workflows. Built support for zero-shot, supervised, and unsupervised embedding modes.

AgentInferKit: Modular Platform for Agent Inference

GitHub →

Built a modular platform for LLM/VLM/Agent systems across text, multimodal, retrieval-augmented generation, and tool-use tasks. Designed core components including model adapters, reasoning strategies, RAG pipelines, and evaluation modules.

OpenLab: Email-Triggered Automated Experiment Runner

GitHub →

Developed an open-source automation system that allows users to submit ML/DL/statistical experiment requests via email, with Claude Code handling experiment planning, code generation, execution, and result packaging.

SOMAS: Online Reinforcement Learning Framework

GitHub →

An online memory and communication-graph optimization framework for multi-agent systems using Swarm and LangGraph. Has earned over 500+ GitHub stars. Improves deployment accuracy by about 40% over existing solutions.

Contact

Get in Touch

15982498982

Wuhan, China