Felt great to attend a local INFORMS annual meeting in Seattle last week! Got to know some very interesting and new research in optimization, energy systems and learning, and got to meet with some old friends.

E-Energy Best Paper Runner-Up

Great honor to get the 2nd Best Paper award at this year’s E-Energy Conference in Phoenix! Trying to advocate more rigorous analysis when applying machine learning/forecasting techniques in power systems. See our paper here!

Also great to attend FCRC 2019! Learned a lot from other venues other than e-Energy such as Sigmetrics, EC and COLT.

Invited Talk at NetSci 2019

Thanks my previous advisor Prof. Yang-Yu Liu at Harvard for the invitation to give a talk at the Netsci conference satellite symposium “Controlling Complex Networks: When Control Theory Meets Network Science”. I showed our works on convex neural nets for control (slides here!). Summer in Vermont felt pretty good, and it felt good to visit many old friends in New England! Also felt honored to talk to groups at University of Vermont.

ACM e-Energy Paper on Vulnerabilities of Load Forecasting Algorithms

Our work Exploiting Vulnerabilities of Load Forecasting Through Adversarial Attacks was accepted at this year’s ACM e-Energy conference. We looked into the overlooked security issues on load forecasting, which is an essential step in power system operations. With only small attack efforts, severe damages could be made upon normal system operations.

Code is available here.

Sincere thanks for the travel grant given by the committee!

Grid Science Winter School

I attended the 3rd Grid Science Winter School & Conference held in Santa Fe by Los Alamos National Laboratory. It was a super interesting venue! Great to hear from a lot of inspiring works, gather with old friends during my summer internship, and showed our work on Optimal Control via Neural Networks as a contributed student talk. Also showed one poster on our ongoing work Data-Driven Vulnerabilities of Power Systems Algorithms.

ICLR Paper on Optimal Control via Neural Networks

Our work on control via neural network (preprint on Arxiv available): Optimal Control Via Neural Networks: A Convex Approach was accepted to ICLR 2019. Very exciting theoretical results on showing a class of neural networks being input-convex, which are able to achieve excellent learning plus control performances on tasks such as Mujoco locomotion and building energy management.

Also nominated as an oral paper for the conference, along with a travel award. Featured in media coverage such as Sohu, 机器之心.