Prediction and Control with Function Approximation
Upgrade your knowledge by doing the third part of the reinforcement learning specialization course offered by the University of Alberta.
Intermediate
Online
5 Weeks
13,514 INR Quick Facts
| Medium Of Instructions | Mode Of Learning | Mode Of Delivery |
|---|---|---|
| English | Self Study | Video and Text Based |
Course Overview
Reinforcement learning is a machine learning method that helps to understand how the software agents should take actions according to the situations and requirements. Helps solve all the problems that have large or high dimensional and potential limitless spaces available. It helps one to investigate how policy prediction and evaluation methods like TD and Monte Carlo can be used to help function approximation settings.
Prediction and Control with Function Approximation certification course by Coursera include features construction techniques for representational learning through a neutral network, RL and backdrop. This course is designed for the candidates who have completed the first and the second part of the reinforcement learning and specialization training as it is a continuation of the same.
By doing the Prediction and Control with Function Approximation training course they will be a field ready and gain knowledge that will help them master their skills and help upgrade their portfolio which will, in the end, help them towards their successful career.
The Highlights
- Approx. 21 hours course
- 100% online learning
- Flexible deadlines
- Favourable for candidates with intermediate knowledge.
- 7 days of free trial available.
- Offered by the University of Alberta
Programme Offerings
- quizzes
- Readings
- assignments
- Projects
Courses and Certificate Fees
| Certificate Availability | Certificate Providing Authority |
|---|---|
| yes | Coursera |
The Prediction and Control with Function Approximation certification fee details :
| Head | Amount |
| 1 month (18 hours/week) | Rs. 6,757 |
| 3 month (6 hours/week) | Rs. 13,514 |
| 6 month (3 hours/week) | Rs. 20,271 |
Eligibility Criteria
Work experience
The candidate should have experience of working with python for at least a year.
Education
The candidates are expected to complete the first two parts of the reinforcement learning specialization course
Certification qualification details
Candidates should successfully attend and complete the entire programme to get the Prediction and Control with Function Approximation certification by Coursera.
What you will learn
After completion of the Prediction and Control with Function Approximation certification syllabus, the candidates will learn to:
- They will be able to start implementing the policy gradient method also called the Actor-Critic method on a discrete state environment
- They will learn the implementation of TD effectively
- They will learn the use of supervised learning approaches using machine learning.
- They will understand all the difficulties that come across while moving to function approximation.
Who it is for
With the Prediction and Control with Function Approximation course, the candidates who have an interest for pursuing careers as Python Programmers, and ML Engineers.
Application Details
Candidates need to follow a set process for the Prediction and Control with Function Approximation classes admission process
Step 1: Candidates need to apply on the programme URL https://www.coursera.org/learn/prediction-control-function-approximation
Step 2: Select the tab Enrol for free. The 7-day free trial will be given once the candidate signs up.
Step 3: As 7 days come to an end, the candidate needs to make a monthly fee paid to access the programme again.
The Syllabus
Videos
- Meet your instructors
- Course 3 Introduction
Readings
- Reinforcement Learning Textbook
- Read Me: Pre-requisites and Learning Objectives
Discussion Prompt
- Meet and Greet
Videos
- Generalization and Discrimination
- Moving to Parameterized Functions
- Framing Value Estimation as Supervised Learning
- Introducing Gradient Descent
- The Value Error Objective
- Gradient Monte for Policy Evaluation
- Semi-Gradient TD for Policy Evaluation
- State Aggregation with Monte Carlo
- Comparing TD and Monte Carlo with State Aggregation
- Week 1 Summary
- The Linear TD Update
- Doina Precup: Building Knowledge for AI Agents with Reinforcement Learning
- The True Objective for TD
- Week 1 Summary
Readings
- Weekly Reading: On-policy Prediction with Approximation
- Module 1 Learning Objectives1
Assignment
On-policy Prediction with Approximation
Programming Assignment
- Semi-gradient TD(0) with State Aggregation
Discussion Prompt
- Good Objectives for Control
Videos
- Generalization Properties of Coarse Coding
- Coarse Coding
- Tile Coding
- What is a Neural Network?
- Using Tile Coding in TD
- Non-linear Approximation with Neural Networks
- Optimization Strategies for NNs
- Gradient Descent for Training Neural Networks
- David Silver on Deep Learning + RL = AI?
- Deep Neural Networks
- Week 2 Review
Readings
- Weekly Reading: On-policy Prediction with Approximation
- Module 2 Learning Objectives
Assignment
Constructing Features for Prediction
Programming Assignment
- Semi-gradient TD with a Neural Network
Discussion Prompt
- Constructing Features for Prediction
Videos
- Episodic Sarsa in Mountain Car
- Episodic Sarsa with Function Approximation
- Expected Sarsa with Function Approximation
- Average Reward: A New Way of Formulating Control Problems
- Exploration under Function Approximation
- Satinder Singh on Intrinsic Rewards
- Week 3 Review
Readings
- Weekly Reading: On-policy Control with Approximation
- Module 3 Learning Objectives
Assignment
Control with Approximation
Programming Assignment
- Function Approximation and Control
Discussion Prompt
- Control with FA #1
- Control with FA #2
Videos
- Advantages of Policy Parameterization
- Learning Policies Directly
- The Objective of Learning Policies
- Estimating the Policy Gradient
- The Policy Gradient Theorem
- Actor-Critic Algorithm
- Gaussian Policies for Continuous Actions
- Demonstration with Actor-Critic
- Congratulations! Course 4 Preview
- Actor-Critic with Softmax Policies
- Week 4 Summary
Readings
- Weekly Reading: Policy Gradient Methods
- Module 4 Learning Objectives
Assignment
Policy Gradient Methods
Programming Assignment
- Average Reward Softmax Actor-Critic using Tile-coding
Discussion Prompt
- Policy Gradient methods
Instructors
University of Alberta, Edmonton Frequently Asked Questions (FAQ's)
The candidate can access the programme for free for 7 days. Hence the fee will not be refunded once paid.
The programme requires 21 hours to complete.
The candidate will learn a series of skills namely, Intelligent Systems, Reinforcement Learning, Function Approximation, Machine Learning and many others.
Yes, the candidates will be awarded certificates once the programme is completed successfully.