Week 4 Lecture 6: Gradient Descent in Machine Learning (PDF)
Document Details
Uploaded by Deleted User
Lakehead University
Saad Bin Ahmed
Tags
Summary
This document is a lecture on gradient descent, focusing on its implementation using Google Colab, presented by Saad Bin Ahmed at Lakehead University. Mathematical formulations, visualizations of results, and discussions on different aspects of learning rate optimization during learning progress are included.
Full Transcript
Revisit-Gradient Descent and its implementation using Google colab Saad Bin Ahmed Department of Computer Science Faculty of Science and Environmental Studies Lakehead University – Thunder Bay 5011-FA - Machine Learning & Neural Nets...
Revisit-Gradient Descent and its implementation using Google colab Saad Bin Ahmed Department of Computer Science Faculty of Science and Environmental Studies Lakehead University – Thunder Bay 5011-FA - Machine Learning & Neural Nets Week 4 Lecture 6 Outline for Lecture 6 Revise the intuition of Gradient Descent Understand Google colab code (Linear Regression.ipynb) Learn implementation of Gradient Descent w.r.t b Learn implementation of Gradient Descent w.r.t b and m (both) in 3D graph Visualization of Learning rate impact using the following link https://developers.google.com/machine-learning/crash-course/fitter/graph Saad B. Ahmed Mathematical Formulation to Calculate m Saad B. Ahmed How search happens to find Best fit line? Saad B. Ahmed Understanding the Learning rate impact Saad B. Ahmed Convergence of Learning curve during iterations Saad B. Ahmed Understanding of Step size Saad B. Ahmed Manipulating Slope and Y-intercept Saad B. Ahmed Impact of Learning rate during learning progression Saad B. Ahmed Slow convergence trend Saad B. Ahmed Starting from smaller value and converging towards minima Saad B. Ahmed Saad B. Ahmed Saad B. Ahmed Few Discussion 1. Effect of Learning Rate 2. The Universality of Gradient Descent Saad B. Ahmed Visualization of Hyperplane Saad B. Ahmed From Simple Linear Regression to Multiple Linear Regression Saad B. Ahmed Saad B. Ahmed