Machine Learning w Neural Nets

Machine Learning w. Neural Nets

Instructor: Dr. Mike Bowles

Neural nets were officially pronounced dead in the 90's, but Canadian scholars didn't get the memo. Their thoughtful, persistence has lead to several new discoveries that have resurrected neural nets and made them the best choice for a number of extremely difficult problems like speech recognition and handwriting recognition. This course will cover background on neural nets, their origins, traditional architectures and recent developments such as auto-encoders and restricted Boltzman machines.

Class will meet for 8 sessions on Wed and Thurs evenings from 7:00 to 9:00 pm.

Meeting Dates
-Mar 27
-Apr 3 and 4
-Apr 11
-Apr 17 and 18
-Apr 24 and 25

Some of the Neural Net Topics that we'll cover
Intro
-What are NN?
-How do they relate to brain?
-Neuron Models

Perceptron
-How perceptron learns
-Properties of perceptron
-Perceptron for reduced rank approximation

Backpropagation methods
-Batch size effects
-Other methods for performing minimization
Recurrent neural nets
How to regularize neural nets
Auto-encoders
Boltzman machines
Deep belief nets
Software packages

Prerequisites
The class will undergraduate level calculus and linear algebra and moderate programming skills. We'll use R and Python primarily. Some familiarity with machine learning problems (e.g. linear regression) will be helpful, but not strictly required.

Registration for the Class
Early Registration - $325
by credit card only - http://machinelearningneuralnets.eventbrite.com
Regular Registration - $375
by credit credit card - http://machinelearningneuralnets.eventbrite.com
by check or cash - in class