Using machine learning for doing rather than knowing:- Practical applications of RL in industry

Today's industrial systems (supply chains, logistics, transportation, energy) usually come with strong real-time sensing capabilities. We can model these distributed systems as a set of signals communicating their health to a control algorithm. However, understanding and managing such systems smoothly is not possible using just supervised and unsupervised learning algorithms for forecasting or anomaly detection. There is a pressing need to develop data-driven algorithms that automatically produce control signals in an adaptive fashion, while operating in non-stationary environments. In this talk, I will talk about the use of reinforcement learning (RL) for achieving this objective. Unlike the more popular versions of RL which focus on gameplay, I will present case studies on real-world industrial operations where RL is used in conjunction with existing rules and constraints to generate higher efficiency.

Harshad is a senior scientist with TCS Research, a visiting associate professor with IIT Bombay, and an associate researcher with RBCDSAI, IIT Madras. His research interest is in control and optimisation of large distributed systems, especially in problems related to operations research. He holds a bachelors degree from IIT Bombay, and masters and PhD degrees from the Massachusetts Institute of Technology.

Using machine learning for doing rather than knowing:- Practical applications of RL in industry

Abstract

Harshad Khadilkar, TCS Research