Back

Speaker "Arvind Rapaka" Details Back

 

Topic

Streamlining the training and productionizing large scale Deep Learning Models

Abstract

Training and operationalization of Deep Learning models that scale to billions of predictions present a complex set of unprecedented engineering challenges. In this talk, I will share the challenges we’ve solved in building AI infrastructure for training and inference in Autonomous Driving, UAVs, large text analysis, etc. The topics include:
1. Distributed ML Training on the large datasets
2. New Batch Optimization methods for effective training such as LAMB
3. Quantization of large ML models for Inference
4. Our learnings while building large scale ML models for autonomous driving, UAVs, large text analysis and medical imaging.

Profile

Arvind Rapaka is VP of Engineering at Techvedika. He founded eComtics - acquired by Techvedika. He has over 14 years of engineering development experience. He has held held leadership roles at Yahoo, Veraz Networks, and GE. He has as published papers in IEEE and other technical journals.