Empowering Time Series Analysis with Large Language Models: A Survey (IJCAI 2024, collaborated with Morgan Stanley-ML Research Department)

Yushan Jiang, Zijie Pan, Xikun Zhang, Sahil Garg, Anderson Schneider, Yuriy Nevmyvaka, Dongjin Song 

TL;DR: We systematically survey existing methods that leverage LLMs for time series analysis, uniquely categorize them into five groups based on the methodology, and discuss their application tasks and domains. We also discuss and highlight future directions that advance time series analysis with LLMs and encourage researchers and practitioner to further investigate this field.

FedSkill: Privacy Preserved Interpretable Skill Learning via Imitation       (KDD 23, collaborated with NEC Labs-DSSS Department) 

Yushan Jiang, Wenchao Yu, Dongjin Song, Lu Wang, Wei Cheng, Haifeng Chen 

TL;DR: We propose a federated interpretable skill acquisition framework via imitation learning, which exploits the segment-level expert demonstrations from multiple clients and produces representative and transferable skills. 

To the best of our knowledge, this is the first work that equips the federated learning framework with a self-explainable structure for downstream tasks.

Structural Knowledge Informed Continual Multivariate Time Series Forecasting (collaborated with Morgan Stanley-ML Research Department)

Zijie Pan*, Yushan Jiang* (Co-first author), Dongjin Song, Fernando Gama, Sahil Garg, Kashif Rasul, Anderson Schneider, Yuriy Nevmyvaka 

TL;DR: Our method tackles continual forecasting over different regimes of multivariate time series (MTS). It jointly optimizes the learned structure toward the existing structural knowledge and forecasting objectives to characterize the dynamic variable dependencies of each regime, where a sample selection mechanism based on maximum entropy is proposed to maximize the temporal coverage of MTS. 

Spatial-Temporal Graph Data Mining for IoT-enabled Air Mobility Prediction  (IEEE IoTJ 2021, supported by U.S. Department of Transportation)

Yushan Jiang, Shuteng Niu, Kai Zhang, Bowen Chen, Chengtao Xu, et al.

TL;DR: We propose the first air traffic prediction paradigm based on spatial-temporal graph modeling. We provide a detailed study based on Airline On-Time Performance Data of the year 2016 (provided by Bureau of Transportation Statistics), and build an air transportation network (via different inductive biases) for 285 U.S. domestic airports with temporal features. Our framework based on spatial-temporal graph neural networks provides accurate joint predictions of the number, average delay, and average taxiing time of departure and arrival flights at each airport.