Ziyi Zhang

I'm currently a Machine Learning Engineer at TikTok, working on next-generation recommender systems for TikTok Mall. Previously, I worked on large-scale time series forecasting to optimize TikTok Shop's supply chain operations.

I earned my Ph.D. in Computer Engineering from Texas A&M University, supervised by Prof. Nick Duffield. I received my B.Eng. in Electronic Information Engineering from the University of Electronic Science and Technology of China (UESTC).

My research focuses on sequence modeling, causality, and foundation models, particularly in recommender systems and time series forecasting.

Email  /  Google Scholar  /  LinkedIn  

Selected Publications

An LLM-assisted framework that automatically discovers interpretable features from systematic forecasting errors to improve frozen pretrained time series forecasters without fine-tuning.

Motivated by marginal improvements on standard benchmarks, accuracy law is proposed to characterize the limits of forecasting accuracy and identify saturated benchmarks in deep time series forecasting.

A unified framework that leverages intervention-induced heterogeneity to jointly recover Granger causal structure, identify unknown interventional targets, and account for latent confounding.

A theoretically guranteed method that infers Granger causal structure and identifies unknown interventional targets by leveraging heterogeneous interventional time series data.

A novel time series clustering framework that effectively handles high dimensionality, noise, and time distortions, enabling better geo-spatiotemporal decision-making.

A reinforcement learning-based method that optimizes evacuation routes by dynamically adapting to real-time disaster conditions.