Duration

2 days

Overview

This hands-on session with Deep Reinforcement Learning will deep dive into the reinforcement learning realm and its architecture. The hands-on session will use the data science stack of the Python libraries commonly used for ML and RL.

The course begins with reinforcement learning systems that emphasizes the successes of approaches found in the fields of neuroscience and animal conditioning research. The course concludes with generalizing much wider problem spaces including sophisticated gameplay, online ad-placement, digital resource management, optimized control systems, and self-driving vehicles.

The course will help one to enlighten themselves about confusions like: what it is, how it works, and how you can apply it to real-world problems.

This 2-day course covers all the basics and advanced concepts about Deep Reinforcement Learning.

Pre-requisites

Python data science stack (Numpy, Pandas, Keras etc.)
Exposure to Machine Learning models using sklearn
Should have good understanding of Regression algorithm
Good to have CNN, LSTM, MLP and Deep Learning knowledge

Intro To Reinforcement Learning

Basic Paradigm
Comparison with other Machine Learning Algorithms
Common Terminology

10:30am - 10:45am Morning Break

10:45am - 12:30pm

Agent
State
Action
Environment
Reward

Markov Decision Process

Optimal Policy
Discretization
Reward augmentation

3:30pm - 3:45pm Afternoon Break

3:45pm - 5:00pm

Value function & Q-value function
Bellman Equation
Q-learning

Q-Network

Basic Architecture
Training The Network - Experience Replay
Deep Q-Learning With Experience Replay

10:30am - 10:45am Morning Break

10:45am - 12:30pm

Policy Gradient

Optimal Policy - Gradient Ascent
Reinforce Algorithm

12:30pm - 1:30pm Lunch Break

1:30pm - 3:30pm

Variance reduction
Actor-Critic Algorithm

3:30pm - 3:45pm Afternoon Break

3:45pm - 5:00pm

Real World Solutions

Recurrent Attention Model
Revision and Summary

Programme scope, delivery arrangements and suitability are confirmed during enquiry.