Hello! 👋

I'm Zixuan Song

Neuroscience, Computer Science, and Mathematics Computational Neuroscience and Machine Learning

About Me

I build end-to-end research systems spanning computational neuroscience, machine learning, and data engineering. My work includes behavior tracking with DeepLabCut and YOLO, large-scale model optimization with XGBoost/Transformers/Mixture-of-Experts, distributed training on IBM LSF and Kubernetes, and quantum-classical workflows in TensorCircuit, with a strong focus on reliability, statistical rigor, and production-ready tooling.

Python C++ R Java JavaScript PyTorch TensorFlow Flask React (Ant Design) Git Docker Linux DeepLabCut YOLO XGBoost Transformers Mixture-of-Experts Hyperopt t-SNE/UMAP Time-Series Analysis RAG Data Engineering MLOps Model Optimization LMM/ANOVA Tukey HSD IBM LSF Kubernetes AWS WebSocket Asynchronous Programming Distributed Systems Statistical Modeling Data Visualization TensorCircuit Quantum ML Metal API

Live Demos

Brain Criticality Demo

https://markso.ng/demo/criticality

Canvas Agent

https://markso.ng/demo/canvas

Experience

Data Analysis and Animal Behavioral Studies in Complex Settings

The Hengen Lab, Washington University in St. Louis

Jan 2024 - PresentJan 2024-Present

Developed and deployed an automated parallel behavior-tracking pipeline using DeepLabCut and YOLO, cutting analysis time from several weeks to 3 days.
Designed a labeling, quality-control, and retraining workflow that automates task assignment, incorporates new annotations, and redistributes workload for YOLO-based animal and object tracking models.
Implemented and evaluated behavioral classification models, including XGBoost, Transformer, Mixture-of-Experts, Random Forest, and Logistic Regression, using Hyperopt for large-scale hyperparameter optimization and model selection.
Designed an unsupervised discovery pipeline using t-SNE, UMAP, and PCA to identify and characterize latent clusters of hunting strategies from high-dimensional behavioral features.
Validated behavioral findings using statistical methods including LMM, ANOVA, Tukey HSD, emmeans, and entropy-based analyses.
Built a simulation model of hunting behavior from observational data and validated it against experimental patterns.
Improved the reliability of experimental hardware and software for sleep-deprivation paradigms, eliminating recurring system failures through iterative redesign and troubleshooting.
Contributed to experimental preparation for in vivo studies, including tetrode construction, animal monitoring, and standardized pre- and post-experiment handling.
Built visualization and reporting tools, including an interactive website, for internal analysis and conference presentations; presented results at NEXTEN 2024.
Currently leading manuscript preparation on the computational characterization of adaptive hunting behavior.

Algorithm Design and Application of the Brain Criticality Hypothesis

The Hengen Lab, Washington University in St. Louis

Jan 2024 - PresentJan 2024-Present

Analyzed neural recordings from Alzheimer’s disease mouse models through the Brain Criticality Hypothesis using DCC, d2, and d_beta metrics, and examined relationships with experimental covariates including sleep deprivation and age.
Contributed to the development and optimization of the lab’s d2 Python implementation, correcting numerical edge cases and improving reliability in previously untested regimes.
Designed and benchmarked a d_beta algorithm for criticality analysis and developed validation procedures for metric computation.
Achieved over 90% accuracy in early disease-onset prediction using criticality-derived features while reducing execution time by more than 99% through algorithmic and implementation-level optimization.
Built an interactive demonstration website for criticality analysis.

Robustness of Information Aggregation in LMSR-Based Prediction Markets

CSE 5106 Multi-Agent Systems, Washington University in St. Louis

Jan 2026 - May 2026Jan 2026-May 2026

Implemented a lightweight LMSR market maker and end-to-end prediction market simulator in Python using NumPy, producing reproducible CSV/JSON outputs and Matplotlib visualizations.
Built a heterogeneous set of trading agents, including informed (Bayesian), risk-averse, noise, momentum/mean-reversion, and adversarial strategies.
Defined robustness and convergence metrics for aggregation quality, including KL divergence, total variation distance, convergence time, and post-attack recovery time.
Added market defenses such as transaction fees, order-size and position caps, cooldowns, and adversary active windows, along with solvency-style safeguards.
Developed a YAML-driven experiment framework with parameter sweeps, multi-seed evaluation, and suite-level summary tables in CSV and Markdown.
Wrote pytest sanity tests and structured the codebase for fast iteration and scalable batch execution.

WashU AI Hackathon Agent Project

WashU AI Hackathon

Oct 2025

Placed Top 5 out of 32 teams in HackWashU 2025 AI Hackathon; awarded a $500 prize.
Built the ReAct AI Agent for reasoning-action workflows in real-time academic assistance.
Developed model routing for OpenAI and implemented an asynchronous backend integrating 22 Canvas LMS tools and 4 vector store tools for RAG uploads.
Built UI/UX in React (Ant Design) for faster navigation and improved usability over Canvas.

Research Trainee (Biol 404)

WashU Neurophysiology Lab

Aug 2024 - Dec 2024Aug 2024-Dec 2024

Designed experiments, hypotheses, and evaluation criteria for electrophysiology studies.
Performed surgical procedures and recorded neural activity under anesthesia supervision.
Analyzed waveform data and co-authored lab manuscripts on findings and future directions.

Research Intern and Teaching Assistant

Tencent Quantum Lab

Sep 2022 - Sep 2023Sep 2022-Sep 2023

Collaborated with an international team of PhD-level researchers, contributing to TensorCircuit through feature development, documentation, and code review.
Developed and validated new TensorCircuit functionality with pytest-based unit tests to strengthen reliability and integration coverage.
Assisted release engineering for macOS Apple Silicon builds by resolving packaging dependencies and implementing a Metal API backend.
Optimized tensor contraction performance in C++ and Python bindings, improving throughput by over 30%.
Implemented error-mitigation algorithms (HAMMER, zero-noise extrapolation) on NISQ hardware for improved stability and fidelity.
Led a 12-member training initiative on quantum computing and quantum-enhanced ML, including seminar and onboarding design.

Research Intern

Tsinglan Bio-Adv Lab

Oct 2022 - Aug 2023Oct 2022-Aug 2023

Researched and synthesized melanin-combined photosensitizers.
Evaluated antibacterial effects with SOSG test analysis and wet-lab workflows.
Applied DNA sequencing, PCR, and cell culture techniques in experimental studies.

Publications

Information preference and information supply efficiency evaluation before, during, and after an earthquake: Evidence from Songyuan, China

2021

Li, S., Peng, X., Pang, R., Li, L., Song, Z., & Ye, H.

International Journal of Environmental Research and Public Health 18, 13070 (2021)

Hunting in the Urban Jungle: Unveiling Complex Predatory Behaviors in Mice Through Ecologically-Inspired Environments

Sep 2024

Schneider, A.*, Chitalia, J.*, Song, Z. (M.)*, & Hengen, K.

Poster presented at NEXTEN 2024, St. Louis, MO (September 16, 2024)

*Denotes co-first authorship

Environmental complexity reveals memory-guided search as the locus of learning in prey capture

Jul 2026

Schneider, A. M., McGregor, J. N., Song, M., Amme, J. L., Zheng, S., Wu, D., Tu, J., Yao, G., Eslinger, E., Chitalia, J., Powers, J., Sinha, V., Dyer, E. L., Levenstein, D., & Hengen, K. B.

bioRxiv preprint 2026.06.28.735138 (2026)