Akhilesh Gotmare

I am a Master's student at the Department of Computer Science at EPFL, Switzerland, where I am working with Prof. Martin Jaggi's Machine Learning and Optimization laboratory for my thesis project. Earlier this year, I interned with Dr. Richard Socher's Metamind team at Salesforce in Palo Alto (Apr - Sept 2018). I am currently working on decoupling backpropagation to enable model parallel training and using novel empirical methods for neural net loss landscape and representational analysis.

Prior to joining EPFL, I completed my undergraduate studies in Electrical Engineering (with a CSE minor) from IIT Gandhinagar. In Summer 2015, I interned with Dr. Nitesh Chawla 's iCeNSA lab at the Univ. of Notre Dame.

Email  /  Google Scholar  /  LinkedIn  /  CV / GitHub

Publications
A Closer Look at Deep Learning Heuristics: Learning rate restarts, Warmup and Distillation
(partial results from this work presented at OPTML Workshop, ICML 2018) Akhilesh Gotmare, Nitish Shirish Keskar, Caiming Xiong and Richard Socher
arXiv

Using Mode Connectivity for Loss Landscape Analysis
Akhilesh Gotmare, Nitish Shirish Keskar, Caiming Xiong and Richard Socher
Workshop on Modern Trends in Nonconvex Optimization for Machine Learning, ICML 2018.
arXiv Poster

Decoupling Backpropagation using Constrained Optimization Methods
Akhilesh Gotmare*, Valentin Thomas*, Johanni Brea and Martin Jaggi
Workshop on Efficient Credit Assignment in Deep Learning and Deep Reinforement Learning, ICML 2018.
OpenReview Poster

Unsupervised robust nonparametric learning of hidden community properties
Mikhail Langovoy, Akhilesh Gotmare, Martin Jaggi, Suvrit Sra
arXiv

Swarm and evolutionary computing algorithms for system identification and filter design: A comprehensive review
Akhilesh Gotmare, Sankha Subhra Bhattacharjee, Rohan Patidar, Nithin V. George
Swarm and Evolutionary Computation, 32, 68-84.
paper

Nonlinear system identification using a cuckoo search optimized adaptive Hammerstein model
Akhilesh Gotmare, Rohan Patidar, Nithin V. George
Expert systems with applications, 42(5), 2538-2546.
paper


Website style cloned from Jon Barron's website