Junping Li

I’m Junping Li, MS in Marine Science with major on ocean intelligence at Shanghai Jiao Tong University and BE in Automation (EECS) at Ocean University of China. I study large model and applications at AI Institute, Shanghai Jiao Tong University, prior to this, I studied intelligent control strategy & deep learning of cross environment vehicle. During my university years, I was awarded Outstanding Student, Outstanding Graduate, Academic Excellence Scholarship and Practice Scholarship.

My research covers model training, inference, agent and applications, such as pre-training/CPT, SFT, RL, RAG, agent system, interaction of multiple models in space, mapping nonlinearity of modal space, and social simulation of multi-agent or model spaces with cognition friction or other social factors.
I have been learning and enjoy acquiring new knowledge, such as AI trend, linguistics, causal inference, cognitive science, game theory and so on, also interested in society, sometimes thinking something interesting: human non Bayesian or non scientific mind but enough natural/effective, fuzzy learning that we in most cases are based on rules and fuzzy logic, with explainable expression and network compatibility.

I hope to combine AI, psychology, linguistics and multiple disciplines to propose new ideas, theories and works, and make contributions to the development of human and society and the world we live in.

Publications

Junping Li, et al. Large model multi-agent social simulation with cognitive friction. in preparation, 2026.

Junping Li, Y Bai, S Wen, et al. Interaction of Multiple Large Models in Representation Space: A Comprehensive Survey. Preprint, 2025. junpingli.com/p/interact_multi_lm_space

Junping Li, H Zhou, D Lu, et al. Nonlinear and reinforcement learning control for motion of hybrid aerial underwater vehicle. Neural Computing and Applications, 2025.

Junping Li, Y Jin, R Hu, et al. Trajectory tracking control of hybrid aerial underwater vehicle subject to wind and wave disturbances. Journal of Intelligent & Robotic Systems, 2024.

Projects

Large Model Multi-Agent Social Interaction with Social Factors

Integrate cognitive dissonance theory in psychology and conflict functionalism in sociology to form the friction coefficient, consist of semantic distance, value conflict, emotional polarity difference or others, for the issues of excessive smoothness and false consensus in social simulations.

SmartPsych: large model smart social science psychology field knowledge, research and academy

llamaindex+BGE+milvus knowledge base, pageindex local, markdown+pdf page number mapping+large model to tree index; Hybrid retrieval of text frequency BM25, dense semantics and sparse word importance vector similarities, and concat attention rerank; Inference retrieval with page source, hierarchical path, and inference trace; Agnet workflow: react and function call, context and memory management, reflection; Input/output optimization: query rewriting, answer quotation.

Interaction of Multiple Large Models in the Perspective of Space

Spatial representation and cross-modal alignment & mapping; Model merging & ensembling in representation space; Multi-agent system, representation space interaction, social simulation and metaphor/evolution; Geometric analysis, interpretability and theory of space and interaction.

Project Large Model Know System Engineering

junpingli.com/proj_lm_kn_sys_eng

System, understanding and code quick check of large model training, inference, applications and projects
Basis: transfomer, machine learning and deep learning, language model, large model Pre, model trends; Training: pre-training/CPT, SFT, distillation, RL; Training & inference optimization: data/device/algorithms, RAM/bandwidth/compute, API, function call, prompt; Application: multimodal, RAG, agent, engineering and project

Intelligent Control Strategy and Deep Learning of Cross Environment Vehicle

Key coefficients of robotics by multivariate analysis and neural networks; Intelligent strategy for the large change of environment; Key factors and conditions of the cross environment(CE) with multiple variables, to critical relations & feasible domains; 3-D space cross model, uncertainty and environment difference key problems; Nonlinear adaptive intelligent fuzzy logic and deep reinforcement learning of CE vehicle with deterministic policy & temporal difference learning.

Skills

• C/C++, Matlab & Simulink, Python, PyTorch, LaTeX, FastAPI, Redis, OpenBayes, Drawio, Transformers, Datasets
• Pre-training/CPT, MoE, Distillation, SFT/instruct fine-tuning, PEFT, LoRA/QLoRA, Unsloth, LlamaFactory, Cold Start, RLHF/RLAIF, RLVR, Reward Model, DPO/PPO/GRPO, TRL, verl, Rubrics as Reward, LLM-as-Judge
• Multimodal, CLIP/SigLIP, LLaVA, Training inference optimization, FlashAttention, Quantization, Distributed parallel, Accelerate, DeepSpeed, vLLM, Concurrency, API, Function call, Prompt engineering, CoT
• RAG, LlamaIndex, FAISS, Milvus, Hybrid retrieval, PageIndex, Context engineering
• Agent, ReAct/Planning/Reflection, Agent Loop, Tool use, MCP, Skill, Multi-Agent architecture, Communication mechanism, Control strategy, LangChain/LangGraph, A2A, Agentic RL, Harness, Self-Evolution

Academic Services

Neural Computing and Applications, ICRA, IROS

Contact

Email: ljp.id [at] sjtu.edu.cn
Web: junpingli.com
Address: Shanghai Jiao Tong University, 800 Dongchuan Road, Shanghai 200240