Research

Selected research projects in AI-driven quantitative investment, from academic prototypes to industry deployment.

Unified learning objective across factors, stock selection, and portfolio construction.

Regime-switching VAE for online factor estimation under changing markets.

Financial news events as primary decision units for trading systems.

AI Quantitative Investment

End-to-End Active Investing (E2EAI), 2022–2023
- Built one of the first end-to-end deep learning frameworks covering factor selection, stock selection, and portfolio construction under a unified objective for active investing.
- Published at ICAIF ‘23; deployed in industry backtests and live trading pipelines.
HireVAE: Online Adaptive Factor Model, 2022–2023
- Developed a hierarchical, regime-switching VAE for online factor estimation that adapts to changing market conditions using only point-in-time information.
- Published at IJCAI 2023; applied to multi-market factor investing research.
Deep Multi-Factor Model, 2022
- Designed a graph-attention deep factor model with industry/market neutralization and interpretable factor composition on a dynamic multi-relational stock graph.
- Presented at NeurIPS 2022 GLInd Workshop; see publication.
Live AI Quant Strategies (CSI 300 / CSI 1000), 2022–2024
- Designed and managed deep learning strategies for large- and mid/small-cap A-share indices at a top-tier asset management firm in mainland China.
- Built end-to-end research platforms, automated model pipelines, and a multi-agent GPT factor discovery framework.
Event-Driven AI Trading & Financial Knowledge Graphs, 2024–Present
- Collaborating with Prof. Jian Guo on Janus-Q (end-to-end event-driven trading via hierarchical-gated reward modeling) and FinKario (event-enhanced financial knowledge graph construction, accepted by ACL 2026).

Jump Detection in Global Equity Markets, 2017–2018 (PolyU)
- Stacked LSTM networks for jump detection across 11 global stock indices; parallel regime-switching framework on high-frequency A-H share data.
- Led to Soft Computing publication (2020).
Volatility Modeling & Risk Management, 2014–2017 (PolyU MPhil)
- Fuzzy and multivariate GARCH methods for stock market volatility forecasting and risk management.
- See MPhil thesis and IJFS paper (2018).
Algorithmic Trading with Copulas, 2014
- Copula-based pairs trading strategy and risk measurement.
- Published in Statistics and Decision.
Deep Learning for Finance (SenseTime Internship), 2018
- TCN and LSTM models for futures prediction; simulated limit-order-book environment for reinforcement learning research.