Brian Axelrod focuses on deep learning representations and loss geometry as a lever for understanding and precisely shaping model outputs.

Publications

Conference and Journal Papers
LieRE: Generalizing Rotary Position Encodings	In submission	PDF	Informal Summary
Near-optimal Approximate Discrete and Continuous Submodular Function Minimization	SODA 2020	PDF	Informal Summary
Sample Amplification: Increasing Dataset Size even when Learning is Impossible	ICML 2020	PDF	Informal Summary
A Polynomial Time Algorithm for Log-Concave Maximum Likelihood via Locally Exponential Families [merged]	NeurIPS 2019	PDF	Informal Summary
An Efficient Algorithm For High-Dimensional Log-Concave Maximum Likelihood [original]		PDF	Informal Summary
Hardness of 2D Motion Planning Under Obstacle Uncertainty	IJRR	PDF	Informal Summary
Hardness of 3D Motion Planning Under Obstacle Uncertainty	WAFR 2018	PDF	Informal Summary
Provably Safe Robot Navigation with Obstacle Uncertainty (Journal Version)	IJRR	PDF	Informal Summary
Provably Safe Robot Navigation with Obstacle Uncertainty (Conference Version)	RSS 2017	PDF	Informal Summary
Reducing FPGA Algorithm Area by Avoiding Redundant Computation	ICRA 2015	PDF	Informal Summary

Brian Axelrod’s masters thesis, done under the supervision of Leslie Pack Kaelbling and Tomás Lozano-Pérez at MIT, can be found here.