ARTIFICIAL NEURAL NETWORKS. Overview EdGeneral concepts Areej:Learning and Training Wesley:Limitations and optimization of ANNs Cora:Applications and.

Slides:



Advertisements
Similar presentations
Artificial Neural Networks
Advertisements

A Brief Overview of Neural Networks By Rohit Dua, Samuel A. Mulder, Steve E. Watkins, and Donald C. Wunsch.
NEURAL NETWORKS Backpropagation Algorithm
Neural networks Introduction Fitting neural networks
Artificial Intelligence 13. Multi-Layer ANNs Course V231 Department of Computing Imperial College © Simon Colton.
CSCI 347 / CS 4206: Data Mining Module 07: Implementations Topic 03: Linear Models.
Artificial Neural Networks - Introduction -
Machine Learning: Connectionist McCulloch-Pitts Neuron Perceptrons Multilayer Networks Support Vector Machines Feedback Networks Hopfield Networks.
Machine Learning Neural Networks
Artificial Neural Networks
Neural Networks Basic concepts ArchitectureOperation.
Prénom Nom Document Analysis: Artificial Neural Networks Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
Chapter 5 NEURAL NETWORKS
November 19, 2009Introduction to Cognitive Science Lecture 20: Artificial Neural Networks I 1 Artificial Neural Network (ANN) Paradigms Overview: The Backpropagation.
Prénom Nom Document Analysis: Artificial Neural Networks Prof. Rolf Ingold, University of Fribourg Master course, spring semester 2008.
Convolutional Neural Networks for Image Processing with Applications in Mobile Robotics By, Sruthi Moola.
1 Introduction to Artificial Neural Networks Andrew L. Nelson Visiting Research Faculty University of South Florida.
Artificial Neural Networks
Classification Part 3: Artificial Neural Networks
Artificial Neural Networks
Artificial Neural Network Theory and Application Ashish Venugopal Sriram Gollapalli Ulas Bardak.
Artificial Neural Nets and AI Connectionism Sub symbolic reasoning.
Neural Networks Ellen Walker Hiram College. Connectionist Architectures Characterized by (Rich & Knight) –Large number of very simple neuron-like processing.
Neural Networks AI – Week 23 Sub-symbolic AI Multi-Layer Neural Networks Lee McCluskey, room 3/10
© Copyright 2004 ECE, UM-Rolla. All rights reserved A Brief Overview of Neural Networks By Rohit Dua, Samuel A. Mulder, Steve E. Watkins, and Donald C.
11 CSE 4705 Artificial Intelligence Jinbo Bi Department of Computer Science & Engineering
Artificial Neural Network Yalong Li Some slides are from _24_2011_ann.pdf.
NEURAL NETWORKS FOR DATA MINING
LINEAR CLASSIFICATION. Biological inspirations  Some numbers…  The human brain contains about 10 billion nerve cells ( neurons )  Each neuron is connected.
Artificial Intelligence Methods Neural Networks Lecture 4 Rakesh K. Bissoondeeal Rakesh K. Bissoondeeal.
1 Chapter 11 Neural Networks. 2 Chapter 11 Contents (1) l Biological Neurons l Artificial Neurons l Perceptrons l Multilayer Neural Networks l 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.
Artificial Neural Networks Bruno Angeles McGill University – Schulich School of Music MUMT-621 Fall 2009.
EE459 Neural Networks Backpropagation
Back-Propagation Algorithm AN INTRODUCTION TO LEARNING INTERNAL REPRESENTATIONS BY ERROR PROPAGATION Presented by: Kunal Parmar UHID:
Over-Trained Network Node Removal and Neurotransmitter-Inspired Artificial Neural Networks By: Kyle Wray.
Neural Networks Demystified by Louise Francis Francis Analytics and Actuarial Data Mining, Inc.
Introduction to Neural Networks Introduction to Neural Networks Applied to OCR and Speech Recognition An actual neuron A crude model of a neuron Computational.
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 LECTURE 1 dr Zoran Ševarac FON, 2015.
Neural Networks 2nd Edition Simon Haykin
Artificial Neural Networks for Data Mining. Copyright © 2011 Pearson Education, Inc. Publishing as Prentice Hall 6-2 Learning Objectives Understand the.
Kim HS Introduction considering that the amount of MRI data to analyze in present-day clinical trials is often on the order of hundreds or.
Neural networks (2) Reminder Avoiding overfitting Deep neural network Brief summary of supervised learning methods.
Feasibility of Using Machine Learning Algorithms to Determine Future Price Points of Stocks By: Alexander Dumont.
Deep Learning Overview Sources: workshop-tutorial-final.pdf
Machine Learning Artificial Neural Networks MPλ ∀ Stergiou Theodoros 1.
Machine Learning Supervised Learning Classification and Regression
Neural Network Architecture Session 2
Artificial Neural Networks
Deep Learning Amin Sobhani.
Randomness in Neural Networks
CSE 473 Introduction to Artificial Intelligence Neural Networks
FUNDAMENTAL CONCEPT OF ARTIFICIAL NETWORKS
CSE P573 Applications of Artificial Intelligence Neural Networks
CSE 473 Introduction to Artificial Intelligence Neural Networks
Artificial Intelligence Methods
CS 4501: Introduction to Computer Vision Training Neural Networks II
CSE 573 Introduction to Artificial Intelligence Neural Networks
Artificial Neural Networks
Neural Networks Geoff Hulten.
Lecture Notes for Chapter 4 Artificial Neural Networks
Neural Networks II Chen Gao Virginia Tech ECE-5424G / CS-5824
Artificial Neural Networks
COSC 4335: Part2: Other Classification Techniques
Neural Networks II Chen Gao Virginia Tech ECE-5424G / CS-5824
Deep Learning Authors: Yann LeCun, Yoshua Bengio, Geoffrey Hinton
PYTHON Deep Learning Prof. Muhammad Saeed.
Overall Introduction for the Lecture
Presentation transcript:

ARTIFICIAL NEURAL NETWORKS

Overview EdGeneral concepts Areej:Learning and Training Wesley:Limitations and optimization of ANNs Cora:Applications and examples

Biological Inspiration

Neuron

Activation Function

Errors Is proportional to

Layers – Famous Problem

Layers – Building Blocks

Layers - Idea

Layers – Solution P1

Layers – Solution P2 3 layers“feedforward”

LEARNING AND TRAINING

Learning methods Supervised learning. Learn to predict an output when given an input. Unsupervised learning. Discover a good internal representation of the input.

Types of Artificial Neural Networks Neural network topologies can vary widely, resulting in differences in architecture, function, and behaviour. Feed- forward ANN. Recurrent ANN.

Feed- forward Neural Networks It is the first and the simplest type of ANNs. This type of organisation is also referred to as bottom-up or top-down. It allows signals to travel in one direction only. From input to output through the hidden layers if there were any. There are no feedback (loops or circles). The output of any layer does not affect that same layer.

Teaching process of multilayer ANN Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Backpropagation algorithm Image source [2]

Recurrent Neural Networks It can be called as feed- back or interactive networks. It have signals traveling in both directions (loops and circles). Feedback networks are powerful and can get extremely complicated. Computations derived from earlier input are fed back into the network, which gives them a kind of memory.

LIMITATIONS & OPTIMISATIONS

Overfitting When performance on training data is high but performance on unseen data is poor. Effectively memorizing training data without learning generalized rules.

Split-sample testing Available test data is split into training set, validation set, and test set. Final weights are those used at stopping point. Image source: [1]

Regularization Reduces overfitting by adjusting the training process. Lx regularization adds a regularization term to the cost function. L2 regularization or weight decay: λ is the regularization parameter. Tends towards smaller overall weights. Less sensitive to noise.

Regularization (cont’) L1 regularization: Tends to maintain most useful weights and shrink others.

Regularization (cont’) Dropout involves training with only a subset of neurons active. For each training run, half of the neurons are randomly chosen to be active, the others unused. Final network is less dependent on any particular input or neuron. More testing data is better for generalization, but expensive. Training data can be artificially increased by adjusting existing data.

Learning slowdown Using Mean-Squared-Error for the cost function can result in very slow learning when the derivative of the output is small. Learning can be initially slow if weight initialisation is far from correct.

Learning slowdown (cont’) The Cross-Entropy cost function eliminates the output derivative from the weight gradients. Weights learn solely based on output value and not slope of output when using sigmoid functions.

Learning slowdown (cont’) Alternative cost function is log-likelihood function combined with softmax output layer. Output represents a probability distribution. Useful for classifiers.

Normalization Scales data so that each input lies in same range, reducing influence of one input feature over another. Speeds up training when the process starts with all input features on same scale. If applied to input in training set then must subsequently be applied to all future input.

Normalization (cont’) ●Z-Score or Gaussian normalization scales by mean (μ) and standard deviation (σ). Min-Max normalization scales input to lie in a fixed range. Sigmoidal normalization reduces the effects of outliers in the input.

Vanishing Gradient Problem Increasing number of hidden layers reduces required complexity of network but makes training harder. Vanishing gradient problem is where the gradient of weights in of early layers are small in multi-layer networks compared to later layers and thus learn slower. Alternative is exploding gradient problem. Unstable gradient problem can be avoided with pre- training. Layers are trained one at a time to produce a compressed form of the input.

APPLICATIONS & EXAMPLES

Application Areas 1. Function approximation 2. Pattern recognition 3.Database mining Problems 1. Car insurance policy – claim frequency 2. Door security -- Fingerprint 3. Medical data mining -- Diagnosing lung cancer

Structure Image source:[1]

Image source:[8] Car insurance policy –claim frequency

Door security--fingerprint Image source:[9]

Image source:[10]

References [1] L. L. Priddy and P.E. Keller (2005). Artificial Neural Networks: An introduction. Vol. 68: SPIE press. [2] M. Bernacki. (2005) Principles of training multi-layer neural network using backpropagation. Available: [3] S. Haykin and N. Network, "A comprehensive foundation," Neural Networks, vol. 2, [4] M. James. (2012), The Triumph of Deep Learning. Available: intelligence/5206-the-triumph-of-deep-learning.html [5] J. McCaffrey. (2014), How To Standardize Data for Neural Networks. Visual Studio Magazine. Available: visualstudiomagazine.com/articles/2014/01/01/how-to-standardize-data-for-neural-networks.aspx [6] M. Nielsen. (2015). Neural Networks and Deep Learning. Available: neuralnetworksanddeeplearning.com/ [7] W. Wu, H. Maier, G. Dandy, and R. May, "Exploring the impact of data splitting methods on artificial neural network models," in International Conference on Hydroinformatics (10th: 2012: Hamburg, Germany), [8] [9] [10]

Thank you