Ch4: Backpropagation (BP)

Slides:



Advertisements
Similar presentations
Artificial Neural Networks
Advertisements

Beyond Linear Separability
NEURAL NETWORKS Backpropagation Algorithm
1 Machine Learning: Lecture 4 Artificial Neural Networks (Based on Chapter 4 of Mitchell T.., Machine Learning, 1997)
1 Neural networks. Neural networks are made up of many artificial neurons. Each input into the neuron has its own weight associated with it illustrated.
Neural Networks  A neural network is a network of simulated neurons that can be used to recognize instances of patterns. NNs learn by searching through.
Machine Learning Neural Networks
Lecture 14 – Neural Networks
Supervised learning 1.Early learning algorithms 2.First order gradient methods 3.Second order gradient methods.
Connectionist models. Connectionist Models Motivated by Brain rather than Mind –A large number of very simple processing elements –A large number of weighted.
Back-Propagation Algorithm
Neural Networks Chapter Feed-Forward Neural Networks.
Multi Layer Perceptrons (MLP) Course website: The back-propagation algorithm Following Hertz chapter 6.
Artificial Neural Networks
CHAPTER 11 Back-Propagation Ming-Feng Yeh.
Image Compression Using Neural Networks Vishal Agrawal (Y6541) Nandan Dubey (Y6279)
Neural Networks. Background - Neural Networks can be : Biological - Biological models Artificial - Artificial models - Desire to produce artificial systems.
Face Recognition Using Neural Networks Presented By: Hadis Mohseni Leila Taghavi Atefeh Mirsafian.
Review – Backpropagation
Artificial Neural Networks
Biointelligence Laboratory, Seoul National University
Multiple-Layer Networks and Backpropagation Algorithms
Artificial Neural Networks
Neural Networks Chapter 6 Joost N. Kok Universiteit Leiden.
Machine Learning Chapter 4. Artificial Neural Networks
Appendix B: An Example of Back-propagation algorithm
NEURAL NETWORKS FOR DATA MINING
Classification / Regression Neural Networks 2
Artificial Intelligence Techniques Multilayer Perceptrons.
CS 478 – Tools for Machine Learning and Data Mining Backpropagation.
Artificial Neural Networks An Introduction. What is a Neural Network? A human Brain A porpoise brain The brain in a living creature A computer program.
1 Lecture 6 Neural Network Training. 2 Neural Network Training Network training is basic to establishing the functional relationship between the inputs.
Neural Networks - lecture 51 Multi-layer neural networks  Motivation  Choosing the architecture  Functioning. FORWARD algorithm  Neural networks as.
Neural Networks Presented by M. Abbasi Course lecturer: Dr.Tohidkhah.
Neural Networks Teacher: Elena Marchiori R4.47 Assistant: Kees Jong S2.22
Neural Networks 2nd Edition Simon Haykin
Neural Networks The Elements of Statistical Learning, Chapter 12 Presented by Nick Rizzolo.
Learning: Neural Networks Artificial Intelligence CMSC February 3, 2005.
Learning with Neural Networks Artificial Intelligence CMSC February 19, 2002.
CSE343/543 Machine Learning Mayank Vatsa Lecture slides are prepared using several teaching resources and no authorship is claimed for any slides.
Neural networks.
Multiple-Layer Networks and Backpropagation Algorithms
Neural Network Architecture Session 2
Artificial Neural Networks
Supervised Learning in ANNs
Deep Learning Amin Sobhani.
Data Mining, Neural Network and Genetic Programming
Neural Networks A neural network is a network of simulated neurons that can be used to recognize instances of patterns. NNs learn by searching through.
Supervised Training of Deep Networks
Structure learning with deep autoencoders
Classification / Regression Neural Networks 2
CSC 578 Neural Networks and Deep Learning
Artificial Neural Network & Backpropagation Algorithm
Synaptic DynamicsII : Supervised Learning
Neuro-Computing Lecture 4 Radial Basis Function Network
Ch2: Adaline and Madaline
network of simple neuron-like computing elements
Artificial Neural Networks
Neural Network - 2 Mayank Vatsa
An Introduction To The Backpropagation Algorithm
Neural Networks Geoff Hulten.
CSSE463: Image Recognition Day 18
Ch4: Backpropagation (BP)
CSSE463: Image Recognition Day 18
Neural networks (1) Traditional multi-layer perceptrons
COSC 4335: Part2: Other Classification Techniques
Computer Vision Lecture 19: Object Recognition III
CS621: Artificial Intelligence Lecture 22-23: Sigmoid neuron, Backpropagation (Lecture 20 and 21 taken by Anup on Graphical Models) Pushpak Bhattacharyya.
Introduction to Neural Networks
Outline Announcement Neural networks Perceptrons - continued
Presentation transcript:

Ch4: Backpropagation (BP) Werbos -> Ponker -> Rummelhart -> McClelland 。BP Architecture: Characteristics: Multilayer, feedforward, fully connected

。 Potential problems being solved by BP 1. Data translation, e.g., data compression 2. Best guess , e.g., pattern recognition, classification Example: Character recognition application a) Traditional method: Translate a 7 × 5 image to 2–byte ASCII code

Lookup table Suffer from: a. Noise, distortion, incomplete b. Time consuming

b) Recent method: Recognition-by-components Traditional approach Neural approach

4.1. BP Neural Network During training, self-organization of nodes on the intermediate layers s.t. different nodes recognize different features or their relationships. Noisy and incomplete patterns can thus be handled.

4.1.2. BP NN Learning Given training examples: , where find an approximation of through learning

。 Learning cycle 7

4.2. Generalized Delta Rule (GDR) Consider input vector Hidden layer: Net input to the jth hidden unit hidden layer jth hidden unit ith input unit bias term with jth unit Output of the jth hidden unit transfer function

Output layer: 。 Update of output layer weights The error at a single output unit k, The error to be minimized: where M: # output units

The descent direction The learning rule: where: learning rate

。 Determine where L: # hidden units 11

The weights on the output layer are updated as 。 Consider Two forms for the output functions i) Linear ii) Sigmoid or 12

For linear function (A) For sigmoid function Let (A) 14

。 Example 1: Quadratic neurons for output nodes Output function: Sigmoid Determine the updating equations of for output-layer neurons. 15

◎ Updates of hidden-layer weights Difficulty: Unknown outputs of the hidden-layer units Idea: Relate error E to the output of the hidden layer 16

17

Consider sigmoid output function 18

19

BPN Summary

※ The known error (or loss) on the output layer are propagated back to a hidden layer of interest to determine the weight changes on that layer

4.3. Practical Considerations 。 Principles of determining network size: i) Use as few nodes as possible. If the NN fails to converge to a solution, it may need more nodes. ii) Prune the hidden nodes whose weights change very little during training 。 Principles of choosing training data i) Cover the entire domain (representative) ii) Use as many data as possible (capacity) iii) Adding noise to the input vectors (generalization)

。 Parameters: i) Initialize weights with small random values ii) Learning rate η decreases with # iterations η small slow; η large perturbation iii) Momentum technique -- Adding a fraction of the preview change, while tends to keep the weight changes going in the same direction to the weight change, iv) Perturbation – Repeat training using multiple initial weights.

4.4. Applications Dimensionality reduction: A BPN can be trained to map a set of patterns from an n-D space to an m-D space (m < n).

Data compression - video images The hidden layer represents the compressed form of the data. The output layer represents the reconstructed form of the data. Each image vector will be used as both the input and the target output.

‧ Size: NTSC: National Television Standard Code 525 × 640 = 336000 #pixels/image ‧ Strategy: Divide images into blocks, e.g., 8 × 8 = 64 pixels,  64-output layer, 16-hidden layer, 64-input layer, #nodes = 144

◎ Paint quality inspection Reflects a laser beam off the painted panel and onto a screen Poor paint: Reflected laser beam diffused ripples, orange peel, lacks shine Good paint: Relatively smooth and bright luster Closely uniform throughout its image

。 Idea

The output was to be a numerical score (1(best) -- 20(worst))