I am currently a second year MSR (master of science in robotics) student at Carnegie Mellon University and honorably
co-advised by Professor Zhihao Jia and
Professor Changliu Liu. Before coming to CMU, I received my B.S.
degree in CS at Renmin University of China where I was honorably advised by Professor
Qin Jin.
My current research interests lie in the intersection of machine learning system and deep learning theory.
Goals. 1. Making well-trained deep models (harder one: during training as well) more computational and storage efficient
from both software and hardware end. 2. Moving a step forward towards knowledge database and reusability of learned knowledge.
Interests. Machine Learning System, Deep Learning Theory
Projects
-
Topology Guided Network Compression (currently working on...)
[code]
Trying to guide multiple network compression techniques by the analysis of topological changing flow within the learned model.
-
GradSign: Model Performance Inference with Theoretical Insights
[pdf]
A key challenge in neural architecture search (NAS) is quickly inferring the predictive performance of a broad spectrum of networks to discover statistically accurate and computationally efficient ones. We refer to this task as model performance inference (MPI). The current practice for efficient MPI is gradient-based methods that leverage the gradients of a network at initialization to infer its performance. However, existing gradient-based methods rely only on heuristic metrics and lack the necessary theoretical foundations to consolidate their designs. We propose GradSign, an accurate, simple, and flexible metric for model performance inference with theoretical insights. The key idea behind GradSign is a quantity Ψ to analyze the optimization landscape of different networks at the granularity of individual training samples. Theoretically, we show that both the network’s training and true population losses are proportionally upper-bounded by Ψ under reasonable assumptions. In addition, we design GradSign, an accurate and simple approximation of Ψ using the gradients of a network evaluated at a random initialization state.
-
Spatio-Temporal Graph Dual-Attention Network for Multi-Agent Prediction and Tracking
[pdf]
Effective understanding of the environment and accurate trajectory prediction of surrounding dynamic obstacles are indispensable for intelligent mobile systems (like autonomous vehicles and social robots) to achieve safe and high-quality planning when they navigate in highly interactive and crowded scenarios. Due to the existence of frequent interactions and uncertainty in the scene evolution, it is desired for the prediction system to enable relational reasoning on different entities and provide a distribution of future trajectories for each agent. In this paper, we propose a generic generative neural system (called Social-WaGDAT) for multi-agent trajectory prediction, which makes a step forward to explicit interaction modeling by incorporating relational inductive biases with a dynamic graph representation and leverages both trajectory and scene context information. We also employ an efficient kinematic constraint layer applied to vehicle trajectory prediction which not only ensures physical feasibility but also enhances model performance.
Publications
-
Spatio-Temporal Graph Dual-Attention Network for Multi-Agent Prediction and Tracking, IEEE Transactions on Intelligent Transportation Systems
Jiachen Li, Hengbo Ma, Zhihao Zhang, Masayoshi Tomizuka
Preprint
-
GradSign: Model Performance Inference with Theoretical Insights, to be appeared on ICLR 2022 main conference
Zhihao Zhang, Zhihao Jia
Teaching
-
2019, Teaching Assistant, Multimedia technology, Renmin University of China