Stylized horse-headed portrait of Zixing Jia

Zixing Jia

Physics x Neuroscience x AI

School of Physics and Astronomy, Sun Yat-sen University Academy for Advanced Interdisciplinary Studies, Peking University Ni Ji's Lab, Chinese Institute for Brain Research, Beijing

I am an undergraduate student at Sun Yat-sen University (Class of 2027), majoring in Physics with a strong interdisciplinary focus on brain-inspired computation and neuro-AI systems. I have been admitted to the PhD program at the Academy for Advanced Interdisciplinary Studies (AAIS), Peking University. My background spans optical interferometry, deep learning, and computational neuroscience.

I am currently a research intern at the Chinese Institute for Brain Research, Beijing (CIBR), working on explicit neural reasoning architectures.

Research Intern @ CIBR Beijing

Physics x Neuroscience x AI

  • 01

    Brain-Inspired AI

    Building next-generation AI systems such as memory-augmented networks and neuro-symbolic reasoners that inherit robustness, efficiency, and generalization from biological intelligence.

  • 02

    AI for Neuroscience

    Developing computational tools and machine learning methods to model neural dynamics, decode brain representations, and bridge biophysical mechanisms with algorithmic principles.

  • 03

    Brain-Inspired World Model

    Building internal world models inspired by hippocampal cognitive maps and predictive representations, enabling agents to simulate, plan, and generalize beyond direct experience.

Journal & Conference

NeurIPS 2026 · Poster

Neural Structural Reasoner: A Brain-inspired Architecture for Reasoning over Structured Knowledge

Z. Jia, Y. Pan, N. Ji

Advances in Neural Information Processing Systems (NeurIPS), 2026.

A brain-inspired architecture that preserves relational structure in layered network connectivity, achieving competitive accuracy with human-readable reasoning traces over knowledge graphs.

Project page Interactive blog Details & figures

NSR integrates hierarchical representation with relation-driven state transitions and Hebbian-type learning. Beyond benchmark accuracy, its inference dynamics natively expose human-readable reasoning traces, and the model spontaneously discovers interpretable latent structure — e.g., projecting nations by their learned relational connections yields four clusters that correspond to Cold-War political blocs.

OpenReview: 5 / 4 / 4  ·  Average rating 4.33  ·  Decision: Accept (Poster)

Published - JCR Q1

Multimodal Fusion Network for Microdisplacement Measurement via Michelson Interferometer

Z. Jia, J. Li, X. Li et al.

Optics and Laser Technology 201 (2026) 115225 (CAS Zone 2 Top Optics Journal).

Developed a dual-head multimodal fusion network (MFN) to resolve lambda/2 phase ambiguity and achieve nanoscale displacement inference.

Details & figures

MFN fuses three interferogram channels (raw / differential / FFT) through a MobileViT-CNN hybrid backbone with an LSTM temporal encoder, then splits into a displacement regression head and an interference-order classification head. It maintains stable precision under severe noise while running over four orders of magnitude faster than the traditional HAA fitting pipeline.

Interferometric Micro-Displacement Measurement Method and System Based on Multimodal Fusion

First Inventor - Application No.: 202511848783.3 - Passed Preliminary Examination

Knowledge Graph Structured Reasoning Method and System Based on Brain-Inspired Hierarchical Neural Networks

In Preparation