Home

by Séb Arnold, March 22, 2018

I am a doctoral student in the Theoretical and Emprical Data Science Lab supervised by Fei Sha. My research topic lies at the intersection of optimization and reinforcement learning.

I did my undergraduate at the University of Southern California, double majoring in Computer Science and Mathematics. My research focused on machine learning, distributed systems, and mathematical optimization.

I also like skiing (a lot).

[Contact, Resume]

Me and motorcycle hair

 




News

WIRED Article on Kleo

Matt Simon from WIRED brilliantly covered our work on Kleo the robotic cat from the ValeroLab ! [Article]


Randopt at PyCon18

Randopt was accepted as a poster presentation to PyCon18. For the occasion, I am revamping the website and documentation. See you in Cleveland ! [Website, wiki]


PyTorch Distributed Tutorial

I just published an official tutorial on the distributed package of PyTorch. Thanks to the community for their help in answering my questions ! [Tutorial]




Publications

Shapechanger: Environments for Transfer Learning

S. Arnold, E. Pun, T. Denisart, F. Valero-Cuevas, 2017, SoCal Robotics Symposium [ArXiv, pdf, website]


Accelerating SGD for Distributed Deep Learning Using an Approximated Hessian Matrix

S. Arnold, C. Wang, 2017, ICLR Workshop [ArXiv, pdf]


A Performance Comparison between TRPO and CEM for Reinforcement Learning

S. Arnold, E. Chu, F. Valero-Cuevas, 2016, SoCal ML Symposium


A Greedy Algorithm to Cluster Specialists

S. Arnold, 2016, ArXiv Pre-prints [ArXiv, pdf]




Projects

Randopt

Randopt is a python package to streamline the search for good hyper-parameters. It provides a programmatic interface to serialize, choose, and visualize hyper-parameters and results. [GitHub Repo, Website Post]

Paperify.py

A tool to convert markdown files into beautiful LaTeX docs, web pages, and presentations. A work in progress. [Web Demo, PDF Examples]

Tooski

Simply the best skiing website ever ? [Website]