Numerical Machine Learning
Syllabus

DESCRIPTION:
This course explores problems in classification/pattern recognition (OCR, speech, vision, fault detection, medical diagnosis), regression/function approximation, robot control, and reinforcement learning. We will use techniques from neural networks, statistics, machine learning, and artificial intelligence.

GOAL:
The goal of this course is to enable you to build systems that do something, rather than encyclopedic coverage.

WHAT IS COVERED:

• Introduction
  • A quick sampling of tasks (classification, regression, and reinforcement learning).
  • A quick sampling of representations.
• Tasks:
  • regression/function approximation
  • classification/pattern recognition
  • clustering
  • density estimation
• Representations:
  • polynomials
  • splines
  • radial basis functions
  • sigmoidal feedforward neural nets
  • projection pursuit regression
  • decision trees
  • memory-based learning
  • nearest neighbor methods
  • locally weighted regression
• Training techniques:
  • gradient descent
  • second order methods
    • diagonalized Hessian methods
    • direction set methods
    • Newton (full Hessian) methods
• Feature selection
• Tuning the representation
• Robot Control
• Static Optimization
• Optimization Over Time
• Reinforcement Learning.

What I probably won't cover:
Hopfield nets
ARTn
BAM
Kohonen nets
Mixtures of experts
Recurrent nets or
 other forms of 
 arbitrarily connected nets. 
Boltzman machines
Fuzzy logic
MARS
CMAC