Extreme Learning Machine

Slides:



Advertisements
Similar presentations
Multi-Layer Perceptron (MLP)
Advertisements

NEURAL NETWORKS Backpropagation Algorithm
EE 690 Design of Embodied Intelligence
Introduction to Neural Networks Computing
CSCI 347 / CS 4206: Data Mining Module 07: Implementations Topic 03: Linear Models.
Classification and Prediction: Regression Via Gradient Descent Optimization Bamshad Mobasher DePaul University.
1cs542g-term High Dimensional Data  So far we’ve considered scalar data values f i (or interpolated/approximated each component of vector values.
Supervised learning 1.Early learning algorithms 2.First order gradient methods 3.Second order gradient methods.
Chapter 5 NEURAL NETWORKS
1 Neural Nets Applications Vectors and Matrices. 2/27 Outline 1. Definition of Vectors 2. Operations on Vectors 3. Linear Dependence of Vectors 4. Definition.
Improved BP algorithms ( first order gradient method) 1.BP with momentum 2.Delta- bar- delta 3.Decoupled momentum 4.RProp 5.Adaptive BP 6.Trinary BP 7.BP.
CHAPTER 11 Back-Propagation Ming-Feng Yeh.
CS 4700: Foundations of Artificial Intelligence
1cs542g-term Notes  Extra class next week (Oct 12, not this Friday)  To submit your assignment: me the URL of a page containing (links to)
Matrices CS485/685 Computer Vision Dr. George Bebis.
11 CSE 4705 Artificial Intelligence Jinbo Bi Department of Computer Science & Engineering
Appendix B: An Example of Back-propagation algorithm
ECE 8443 – Pattern Recognition ECE 8527 – Introduction to Machine Learning and Pattern Recognition LECTURE 16: NEURAL NETWORKS Objectives: Feedforward.
Artificial Intelligence Methods Neural Networks Lecture 4 Rakesh K. Bissoondeeal Rakesh K. Bissoondeeal.
CS 478 – Tools for Machine Learning and Data Mining Backpropagation.
From Machine Learning to Deep Learning. Topics that I will Cover (subject to some minor adjustment) Week 2: Introduction to Deep Learning Week 3: Logistic.
A note about gradient descent: Consider the function f(x)=(x-x 0 ) 2 Its derivative is: By gradient descent. x0x0 + -
Multi-Layer Perceptron
11 1 Backpropagation Multilayer Perceptron R – S 1 – S 2 – S 3 Network.
Back-Propagation Algorithm AN INTRODUCTION TO LEARNING INTERNAL REPRESENTATIONS BY ERROR PROPAGATION Presented by: Kunal Parmar UHID:
1  Problem: Consider a two class task with ω 1, ω 2   LINEAR CLASSIFIERS.
Neural Networks - lecture 51 Multi-layer neural networks  Motivation  Choosing the architecture  Functioning. FORWARD algorithm  Neural networks as.
CHAPTER 10 Widrow-Hoff Learning Ming-Feng Yeh.
Review of Matrix Operations Vector: a sequence of elements (the order is important) e.g., x = (2, 1) denotes a vector length = sqrt(2*2+1*1) orientation.
Instructor: Mircea Nicolescu Lecture 8 CS 485 / 685 Computer Vision.
BACKPROPAGATION (CONTINUED) Hidden unit transfer function usually sigmoid (s-shaped), a smooth curve. Limits the output (activation) unit between 0..1.
Neural Networks The Elements of Statistical Learning, Chapter 12 Presented by Nick Rizzolo.
Lecture 2 Introduction to Neural Networks and Fuzzy Logic President UniversityErwin SitompulNNFL 2/1 Dr.-Ing. Erwin Sitompul President University
Learning: Neural Networks Artificial Intelligence CMSC February 3, 2005.
Learning with Neural Networks Artificial Intelligence CMSC February 19, 2002.
Combining Models Foundations of Algorithms and Machine Learning (CS60020), IIT KGP, 2017: Indrajit Bhattacharya.
Neural networks.
MAT 322: LINEAR ALGEBRA.
Fall 2004 Backpropagation CS478 - Machine Learning.
Deep Feedforward Networks
CS479/679 Pattern Recognition Dr. George Bebis
Review of Matrix Operations
Learning with Perceptrons and Neural Networks
One-layer neural networks Approximation problems
第 3 章 神经网络.
CSE 473 Introduction to Artificial Intelligence Neural Networks
Announcements HW4 due today (11:59pm) HW5 out today (due 11/17 11:59pm)
LECTURE 28: NEURAL NETWORKS
CS621: Artificial Intelligence
Singular Value Decomposition
Machine Learning Today: Reading: Maria Florina Balcan
CSC 578 Neural Networks and Deep Learning
ECE 471/571 - Lecture 17 Back Propagation.
Biological and Artificial Neuron
Biological and Artificial Neuron
CS485/685 Computer Vision Dr. George Bebis
Orthogonality and Least Squares
Biological and Artificial Neuron
network of simple neuron-like computing elements
Artificial Neural Networks
What can we know from RREF?
Backpropagation.
LECTURE 28: NEURAL NETWORKS
Artificial Neural Networks
Artificial Intelligence 10. Neural Networks
Backpropagation.
CS621: Artificial Intelligence Lecture 18: Feedforward network contd
Orthogonality and Least Squares
Pattern Recognition: Statistical and Neural
CS621: Artificial Intelligence Lecture 17: Feedforward network (lecture 16 was on Adaptive Hypermedia: Debraj, Kekin and Raunak) Pushpak Bhattacharyya.
Presentation transcript:

Extreme Learning Machine Abdulrhman Aljouie CS698 – Current topics in Data Science 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Outline Extreme Learning Machine Motivation. Structure of the proposed Extreme Learning Machine (ELM). Basis of Extreme Learning Machine. Gradient based Algorithms for Single Hidden Neural Net. Show an Example of training small data using ELM algorithm. Show an Example of predicting small data using ELM algorithm. Matlab functions for ELM Learning and predicting on example data. Extreme Learning Machine Performance Results. 2/23/17 CS698-Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Motivation Feedforward Neural Nets such as back-propagation are slow and the reason is: Training using gradient based learning algorithms e.g. many iterative steps required to achieve better performance. The weights are adjusted iteratively using gradient methods. Extreme Learning Machine (ELM) was proposed to overcome these issues and offer better generalization performance. Gradient based algorithms are slow because: 1. Many iterative learning steps to gain better performance. And other issues such as converge to silly local minima. 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

The proposed algorithm - Extreme Learning Machine (ELM)   Better generalization because if the updates are not adjusted it has been shown in practice that it gives better generalization performance. Also, if the loss function minimize the magnitude of the norm it gives better generalization. 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Structure of Extreme Learning Machine 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Initial idea   To train the network simply find the least square estimate beta hat of linear system HB = T 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Theorem Proof: Tamura and M. Tateishi (1997) 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Issues   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

The # hidden layer units << the # input units 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

H is non-square or singular   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

To calculate the pseudoinverse   If one or more columns, in case columns > rows, are linearly dependent on other cols, e.g. when we add two cols we derive the thirds col. Then the mat is not in full rank. In case mat is square we can say it is linearly independent iff its determinant is non-zero. 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

How to compute the pseudoinverse in case matrix H is not invertible i How to compute the pseudoinverse in case matrix H is not invertible i.e. non-square and H not full rank?   Note: U transpose U is 1 since U is an orthogonal matrix. Eigenvalues are the entries of S and S is Diagonal 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

    2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Extreme Learning Machine                 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Algorithm 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Gradient based Algorithms with Single Hidden Layer Network   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Training Example   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Training Example (continued)   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Training Example (continued)   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Training Example (continued)   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

  2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Classify Example   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Classify Example   2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Matlab code for ELM on the example data 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Training model in Matlab 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

Predicting with function predict 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM have better generalization compared to gradient based algorithm such as BP Extreme Learning Machine (ELM) Gradient based Algorithms e.g. Backpropagation Unique minimum solution Prone to local minima convergence trap Reach both smallest training error and smallest weight norm. Minimizes Error alone. Does not need a stopping methods. Overtraining when objective function doesn’t have proper stopping methods and validation. 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Results on medical application 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine

ELM Results (continued) 2/23/17 CS698 Current Topics in Data Science- Abdulrhman Aljouie Extreme Learning Machine