Chap5.1-NNintro

Chap5.1-NNintro - Machine Learning Neural Networks...

Info iconThis preview shows pages 1–6. Sign up to view the full content.

View Full Document Right Arrow Icon
Machine Learning Srihari Neural Networks Introduction Sargur Srihari
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari Topics 1. Introduction 1. Extending linear models 2. Feed-forward Network Functions 3. Weight-space symmetries 2. Network Training 1. Parameter Optimization 2. Local quadratic approximation 3. Use of gradient information 4. Gradient descent optimization 3. Error Backpropagation 4. The Hessian Matrix 5. Regularization in Neural Networks 6. Mixture Density Networks 7. Bayesian Neural Networks 2
Background image of page 2
Machine Learning Srihari A Neural Network 3 Can be viewed as a generalization of linear models y k (x,w) = σ w kj (2) j = 1 M h w ji (1) i = 1 D x i + w j 0 (1) + w k 0 (2)
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Machine Learning Srihari Linear Models Linear Models for Regression and ClassiFcation have form where x is a D -dimensional vector ϕ j (x) are Fxed nonlinear basis functions e.g., Gaussian, sigmoid or powers of x ±or Regression f is identity function ±or ClassiFcation f is a nonlinear activation function If f is sigmoid it is called logistic regression y (x,w) = f w j φ j (x) j = 1 M f ( a ) = 1 1 + e a Linear Regression Generalized Linear Regression
Background image of page 4
Srihari Extending Linear Models Linear models have limited applicability Due to curse of dimensionality E.g., no of polynomial coeffts needed, Fnding means of Gaussians Extend scope by adapting basis functions ϕ j to data Become useful in large scale problems Both SVMs and Neural Networks address this limitation SVM Varying number of basis functions M centered on training data points Select subset of these during training Neural Network Number of basis functions M Fxed in advance But the ϕ j have their own parameters {w ji } Adapt all parameter values during training 5 y (x,w) = f w j φ j (x) j = 1 M y k (x,w) = σ w kj (2) j
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 6
This is the end of the preview. Sign up to access the rest of the document.

This document was uploaded on 02/25/2012.

Page1 / 18

Chap5.1-NNintro - Machine Learning Neural Networks...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online