Joe Jiang
Wenjie (Joe) Jiang currently works in Google Brain team, where he is focused on applying ML to system design problems. He also worked in network optimization and analysis team, where he applied a variety of mathematical techniques to solving the network capacity planning and demand forecast problem. He received a B.S. from USTC China; M.Phil from Chinese University of Hong Kong; and Ph.D. from Princeton University, all in Computer Science.
Authored Publications
Sort By
Learning Semantic Representations to Verify Hardware Designs
Shobha Vasudevan
Rishabh Singh
Hamid Shojaei
Richard Ho
Thirty-fifth Conference on Neural Information Processing Systems (NeurIPS) (2021)
Preview abstract
We introduce Design2Vec, a representation learning approach to learn semantic abstractions of hardware designs at the Register Transfer Level (RTL). The key idea of our approach is to design a graph convolution based neural architecture that embeds RTL syntax and semantics. We train the architecture on the task of predicting coverage in the design, given some input test stimulus. We then present an approach to use the learnt RTL representation to automatically generate new tests for unseen coverage locations in the design. Our experimental results demonstrate that Design2Vec outperforms several baseline approaches that do not incorporate the RTL semantics and it can be used to generate instantaneous coverage predictions compared to nightly simulation times. Moreover, the tests generated using Design2Vec result in coverage of design points that are difficult to cover for design verification experts using the current manual approaches for test generation.
View details
Efficient Imitation Learning with Local Trajectory Optimization
Jialin Song
Anna Darling Goldie
Navdeep Jaitly
Azalia Mirhoseini
ICML 2020 Workshop on Inductive Biases, Invariances and Generalization in RL (2020)
Preview abstract
Imitation learning is a powerful approach to optimize sequential decision making policies from demonstrations. Most strategies in imitation learning rely on per-step supervision from pre-collected demonstrations as in behavioral cloning or from interactive expert policy queries such as DAgger. In this work, we present a unified view of behavioral cloning and DAgger through the lens of local trajectory optimization, which offers a means of interpolating between them. We provide theoretical justification for the proposed local trajectory optimization algorithm and show empirically that our method, POLISH (Policy Optimization by Local Improvement through Search), is much faster than methods that plan globally, speeding up training by a factor of up to 14 in wall clock time. Furthermore, the resulting policy outperforms strong baselines in both reinforcement learning and imitation learning.
View details
Capacity planning for the Google backbone network
Ajay Kumar Bangla
Ben Preskill
Christoph Albrecht
Emilie Danna
Xiaoxue Zhao
ISMP 2015 (International Symposium on Mathematical Programming) (to appear)
Preview abstract
Google operates one of the largest backbone networks in the world. In this
talk, we present optimization and simulation techniques we use to design the
network topology and provision its capacity to achieve conflicting objectives
such as scale, cost, availability, and latency.
View details