Xin Du · 杜鑫
Menu

@lab

Attractor Lab logo

@ Tongji University · Shanghai Research Institute for Intelligent Autonomous Systems

Shanghai Research Institute for Intelligent Autonomous Systems ｜ Tongji University

Related links → Tongji University Intelligent Robotics and Computational Perception Laboratory · Waseda Language Intelligence Laboratory

What We Study

@lab studies the dynamics, complexity, and semantic compression of foundation models and agents. Our technical framework combines state-space geometry, phase transitions and criticality, information bottlenecks, and generative distributions.

We seek measurable order parameters and distortion functions, then use them in training diagnostics, generation monitoring, agent control, document clustering, and knowledge retrieval.

Research

Five directions share one question: how can macroscopic structure be identified, represented, and controlled from local probabilities, finite states, and heterogeneous observations?

Research Topics

Prospective Students

2027 Master’s Intake

Applications are welcome from prospective master’s students interested in foundation models and agents, complex systems, semantic compression, or semantic models of financial markets.

2027 PhD Intake

Prospective doctoral students, including direct-entry applicants, may work on questions such as:

  • Why do agents become unstable, and how can they operate reliably over long horizons?
  • How can macroscopic complexity and emergent structure in large language models be measured?
  • How do models compress knowledge while retaining task-relevant information under finite capacity?

Research Activities

  • model long-term states, instability mechanisms, and critical transitions in foundation models and agents
  • develop online diagnostics, state regulation, and training methods for generative processes
  • study information compression, generative clustering, knowledge indexing, and retrieval
  • pursue foundational work across AI, complex systems, and information theory

Preparation

  • sustained interest in foundational questions in large models or AI
  • background in linear algebra, probability, or optimization
  • programming experience with Python or PyTorch
  • willingness to undertake long-term research

Research Internships

  • third-year undergraduate students or above; applicants based in Shanghai are preferred
  • interest in foundation models or agent systems
  • availability of at least two days per week
  • remote or on-site work at Tongji’s Zhangjiang campus

Research Environment

  • research-project support and student stipends
  • Tongji University GPU computing platform
  • research offices and accommodation at the Zhangjiang campus

How to Apply

Please send the following materials to duxin@tongji.edu.cn:

  • curriculum vitae
  • academic transcript
  • brief statement of research interests (optional)

Shanghai Research Institute for Intelligent Autonomous Systems

Shanghai Research Institute logo

The institute was established by the Ministry of Education at Tongji University. Its research platforms include:

  • the National Key Laboratory of Autonomous Intelligent Unmanned Systems
  • an NSFC Basic Science Center
  • a Ministry of Education Frontiers Science Center
  • the National AI Industry–Education Integration Innovation Platform

Discipline and Training

  • among China’s first doctoral programs in the interdisciplinary field of Intelligent Science and Technology
  • interdisciplinary training across AI, complex systems, and engineering systems

Zhangjiang Research Campus

The campus covers 43 mu, with a total floor area of 127,500 square metres.

Tongji Zhangjiang campus

Computing Infrastructure

The large-scale GPU computing centre became operational in 2025 and includes more than 20 NVIDIA DGX systems and an agent-computing platform.

GPU computing infrastructure GPU computing infrastructure