About me

Welcome! I am Victor (Yuzhang) Huang, a PhD student in Computer Science at the University of California, Davis, working with Dr. Dipak Ghosal. My research focuses on time series analysis, anomaly detection, and applying multimodal large language models (MLLM) to temporal forecasting problems.

My earlier work covered time series forecasting for finance and other applications like water level and electricity prediction, anomaly detection in networked systems, and sensor calibration. I built production systems for reinforcement learning-based trading, air quality sensor modeling, and wildfire detection using multi-modal data. These projects taught me how real-world temporal data can be non-stationary, have distribution shifts, and require careful modeling choices.

My current research explores how MLLM-based multi-agent systems can help with time series forecasting tasks. I'm interested in whether language models can help automate parts of the experimental workflow—for example, reading related papers, checking code for bugs, running experiments with different configurations, and summarizing results. This includes work on inference-time reasoning and coordination between specialized agents.

I completed my Bachelor of Science in Mathematics at the University at Buffalo, SUNY in 2023.

Research Interests

  • time series icon

    Time Series Analysis and Forecasting

    I study methods for modeling temporal patterns in real-world data, with attention to non-stationarity, distribution shifts, and practical deployment challenges.

  • anomaly detection icon

    Anomaly Detection in Networked Systems

    I work on methods to identify unusual patterns in temporal data from networked environments, focusing on interpretability and robustness to non-stationary conditions.

  • multi-agent icon

    Multi-Agent Reasoning for Temporal Data

    I explore how multiple specialized agents can work together on time series problems, including automating parts of the experimental workflow like running tests, checking results, and comparing different approaches.

  • llm icon

    MLLM-Based Experimental Agents

    I investigate how multimodal large language models can assist with scientific workflows, including inference-time reasoning strategies and model optimization for deployment.

Daniel lewis
quote icon

Daniel lewis

Richard was hired to create a corporate identity. We were very pleased with the work done. She has a lot of experience and is very concerned about the needs of client. Lorem ipsum dolor sit amet, ullamcous cididt consectetur adipiscing elit, seds do et eiusmod tempor incididunt ut laborels dolore magnarels alia.

Resume

Summary: Ph.D. student in Computer Science at UC Davis focusing on time series analysis, anomaly detection, and applying MLLM-based multi-agent systems to temporal forecasting. I work on automating experimental workflows, inference-time reasoning, and coordination between specialized agents. I have experience building scalable ML systems for production environments, with expertise in training/inference optimization, model deployment (PyTorch, vLLM, ONNX, TensorRT), and applications in sensor networks, financial systems, and environmental monitoring.

Education

  1. University of California, Davis

    Sept. 2023 — Expected June 2028

    Doctor of Philosophy in Computer Science
    Research Focus: Time Series Modeling, Network Anomaly Detection, Multi-Agent AI Systems, MLLM
    Advisor: Dr. Dipak Ghosal

  2. University at Buffalo, SUNY

    Aug. 2019 — June 2023

    Bachelor of Science in Mathematics
    Focus: Applied Mathematics, Data Science, Machine Learning

Experience

  1. Graduate Student Researcher

    Dipak Ghosal's Lab, UC Davis
    May 2024 — Present
    • • Conduct research on multi-agent time series modeling, developing frameworks to model long-range dependency (Hurst exponent) and agent coordination.
    • • Designed a multi-agent multi-resolution framework where temporal and spectral models (Transformer, ViT with wavelet representations) act as specialized agents with adaptive routing.
    • • Built a controlled synthetic data pipeline (200K+ samples) and multi-seed training system, enabling reproducible large-scale benchmarking and reducing variance by 30–40%.
    • • Developed coordination strategies across temporal regimes, improving robustness under distribution shifts and heterogeneous time-series patterns.

  2. Machine Learning Engineer Intern

    Interlink Electronics Inc., Fremont, CA
    Feb. 2025 — Sept. 2025
    • • Developed learning-based calibration systems for low-cost air quality sensors, modeling CO and O3 signals under varying environmental conditions (temperature, humidity), improving reliability in real-world deployment.
    • • Built scalable time-series data pipelines for high-frequency sensor streams with automated outlier detection and feature engineering, improving data quality by 20%.
    • • Implemented low-latency, edge-oriented inference pipelines with Docker-based deployment and AWS S3 integration, enabling scalable real-time sensor calibration under resource constraints.

  3. Machine Learning Software Engineer

    Tradnomic Inc., San Jose, CA
    Sept. 2023 — Feb. 2025
    • • Developed a reinforcement learning-based trading system for sequential decision-making (Buy, Sell, Hold) over financial time-series data, modeling market interactions under noisy and non-stationary conditions.
    • • Designed state representations and reward functions based on rolling market features and realized PnL, enabling stable policy learning and profitable strategy optimization.
    • • Built scalable training and deployment pipelines with MLflow, Docker, and Kubernetes, enabling reproducible experiments and distributed inference in production environments.

Research

  1. Adaptive Multi-Agent Multi-Modal Learning for Wildfire Detection

    CHPS Research Lab
    Ongoing Research Project
    • • Designed a multi-agent perception framework where RGB and thermal modalities act as specialized agents, each trained to capture complementary visual patterns under different environmental conditions.
    • • Developed agent-specific training pipelines with modality-aware data augmentation and optimization, improving robustness under distribution shifts such as heavy smoke and low visibility.
    • • Implemented confidence-aware coordination mechanisms, where a coordinator module dynamically aggregates agent predictions based on reliability, enabling adaptive decision-making under sensor degradation.
    • • Built a low-latency inference system using ONNX/TensorRT and INT8 quantization, achieving 2.5× speedup and ≤30ms latency for real-time deployment.

My skills

  • Multi-Agent Systems & Reinforcement Learning
    95%
  • Time Series Modeling & Transformers
    95%
  • PyTorch & Deep Learning Frameworks
    95%
  • Model Optimization (vLLM, ONNX, TensorRT)
    90%
  • Docker, Kubernetes, MLflow
    90%
  • Python Programming
    95%
  • C++ & Systems Programming
    85%
  • SQL & AWS (S3, Ray)
    85%