Lecture notes from "Numerical Analysis for Artificial Intelligence" course I presented at UCSD, CSE department
Switch branches/tags
Nothing to show
Clone or download
Latest commit 5523b49 Oct 23, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
jupyter_notebooks lecture notes Oct 23, 2018
README.md Update README.md Oct 23, 2018
week1_slides.pdf lecture notes Oct 23, 2018
week2_slides.pdf lecture notes Oct 23, 2018
week3_slides.pdf lecture notes Oct 23, 2018
week4_slides.pdf lecture notes Oct 23, 2018
week5_slides.pdf lecture notes Oct 23, 2018

README.md

Numerical Analysis for Artificial Intelligence, Jacek Cyranka

This repository contains course materials from the "Numerical Analysis for Artificial Intelligence" course I presented at UCSD, CSE department during Summer session 2018.

course syllabus

  • Week 1,2 Review of Programming in Python+NumPy+IPython notebook and �Calculus and Linear Algebra topics

    • Python language basics,
    • Linear Algebra in NumPy,
    • Working with Jupyter notebooks,
    • Example problem of solving a linear regression analytically,
    • Functions,
    • Vector spaces,
    • Matrices,
    • Matrix times vector/matrix operation,
    • Matrix transpose/inverse,
    • Solving systems of linear equations,
    • Basic properties,
    • Partial Derivatives,
    • Critical points,
    • Chain rule and gradients,
    • Characterization of critical points as local/global minima/maxima.
  • Week 3 Gradient descent and convex optimization

    • Backpropagation algorithm,
    • gradient checking of a backpropagation implementation,
    • Avoiding problems with convergence by decreasing the learning rate,
    • Accelerated gradient descent (Nesterov momentum method),
    • Minimizing a quadratic function,
    • Solving linear regression using gradient descent.
  • Week 4,5 Nonconvex optimization : supervised learning of feed-forward Neural Networks

    • Difference in Convex/Nonconvex optimization,
    • Classical Blum/Rivest proof that training a 3-node NN is NP-Complete,
    • Perceptrons
    • Single hidden layered networks,
    • Linear, ReLU , tangential networks,
    • Mean squared error loss, cross entropy loss,
    • Multiple hidden layered feed forward networks,
    • Newton’s method.