Presentation is loading. Please wait.

Presentation is loading. Please wait.

Lecture 2 Introduction to Neural Networks and Fuzzy Logic President UniversityErwin SitompulNNFL 2/1 Dr.-Ing. Erwin Sitompul President University

Similar presentations


Presentation on theme: "Lecture 2 Introduction to Neural Networks and Fuzzy Logic President UniversityErwin SitompulNNFL 2/1 Dr.-Ing. Erwin Sitompul President University"— Presentation transcript:

1 Lecture 2 Introduction to Neural Networks and Fuzzy Logic President UniversityErwin SitompulNNFL 2/1 Dr.-Ing. Erwin Sitompul President University http://zitompul.wordpress.com 2014

2 President UniversityErwin SitompulNNFL 2/2 Weights, need to be determined Biological neuron Artificial neuron Bias, need to be determined Learning ProcessesNeural Networks Biological and Artificial Neuron

3 President UniversityErwin SitompulNNFL 2/3 Learning ProcessesNeural Networks Application of Neural Networks Function approximation and prediction Pattern recognition Signal processing Modeling and control Machine learning

4 President UniversityErwin SitompulNNFL 2/4 Building a Neural Network Select Structure: design the way that the neurons are interconnected. Select weights: decide the strengths with which the neurons are interconnected. Weights are selected to get a “good match” of network output to the output of a training set. Training set is a set of inputs and desired outputs. The weight selection is conducted by the use of a learning algorithm. Learning ProcessesNeural Networks

5 President UniversityErwin SitompulNNFL 2/5 Stage 1: Network Training Training Data Stage 2: Network Validation Artificial neural network Input and output sets, adequate coverage Learning Process In the form of a set of optimized synaptic weights and biases Unseen Data From the same range as the training data Artificial neural network Implementation Phase Learning ProcessesNeural Networks Learning Process Knowledge Output Prediction

6 President UniversityErwin SitompulNNFL 2/6 Learning Process Learning is a process by which the free parameters of a neural network are adapted through a process of stimulation by the environment in which the network is embedded. In most cases, due to complex optimization plane, the optimized weights and biases are obtained as a result of a number of learning iterations. [w,b][w,b] x y [w,b] 0 x y(0) Initialize: Iteration (0) [w,b] 1 x y(1) Iteration (1) [w,b] n x y(n) ≈ d Iteration (n) ANN d : desired output … Learning ProcessesNeural Networks

7 President UniversityErwin SitompulNNFL 2/7 Learning Rules Learning ProcessesNeural Networks Error Correction Learning Delta Rule or Widrow-Hoff Rule Memory Based Learning Nearest Neighbor Rule Hebbian Learning Synchronous activation increases the synaptic strength Asynchronous activation decreases the synaptic strength Competitive Learning Boltzmann Learning

8 President UniversityErwin SitompulNNFL 2/8 wk1(n)wk1(n) x1x1 x2x2 xmxm Inputs Synaptic weights Bias Activation function wk2(n)wk2(n) w km (n)  Output y k (n) Desired output d k (n) ek (n)ek (n)  f (.) bk(n)bk(n) 1  Error signal Learning ProcessesNeural Networks Error-Correction Learning  Learning Rule

9 President UniversityErwin SitompulNNFL 2/9 Learning ProcessesNeural Networks Delta Rule (Widrow-Hoff Rule) Minimization of a cost function (or performance index)

10 President UniversityErwin SitompulNNFL 2/10 w kj (0) = 0 y k (n) = [w kj (n) x j (n)] w kj (n+1) = w kj (n) + [d k (n) – y k (n)] x j (n)  : learning rate, [0…1] n = n+1 n = 0 Least Means Squares Rule Learning ProcessesNeural Networks Delta Rule (Widrow-Hoff Rule)

11 President UniversityErwin SitompulNNFL 2/11 Learning ProcessesNeural Networks Learning Paradigm  ANN Error Desired Actual   Environment (Data) Teacher (Expert) Supervised Unsupervised Environment (Data) Delay ANN Delayed Reinforcement Learning Cost Function

12 President UniversityErwin SitompulNNFL 2/12 Single Layer PerceptronsNeural Networks Single Layer Perceptrons Output unit is independent of the others. Analysis can be limited to single output perceptron. Single-layer perceptron network is a network with all the inputs connected directly to the output(s).

13 President UniversityErwin SitompulNNFL 2/13 Single Layer PerceptronsNeural Networks Derivation of a Learning Rule for Perceptrons w1w1 w2w2 E(w)E(w) Key idea: Learning is performed by adjusting the weights in order to minimize the sum of squared errors on a training. Weights are updated repeatedly (in each epoch/iteration). Sum of squared errors is a classical error measure (e.g. commonly used in linear regression). Learning can be viewed as an optimization search problem in weight space.

14 President UniversityErwin SitompulNNFL 2/14 Single Layer PerceptronsNeural Networks Derivation of a Learning Rule for Perceptrons The learning rule performs a search within the solution's vector space towards a global minimum.  The error surface itself is a hyper-paraboloid but is seldom as smooth as is depicted below.  In most problems, the solution space is quite irregular with numerous pits and hills which may cause the network to settle down in a local minimum (not the best overall solution).  Epochs are repeated until stopping criterion is reached (error magnitude, number of iterations, change of weights, etc).

15 President UniversityErwin SitompulNNFL 2/15 Single Layer PerceptronsNeural Networks Derivation of a Learning Rule for Perceptrons Widrow [1962] x1x1 x2x2 xmxm wk1wk1 wk2wk2 w km......  Adaline (Adaptive Linear Element) Goal:

16 President UniversityErwin SitompulNNFL 2/16 Least Mean Squares (LMS) Single Layer PerceptronsNeural Networks The following cost function (error function) should be minimized:

17 President UniversityErwin SitompulNNFL 2/17 Single Layer PerceptronsNeural Networks Least Mean Squares (LMS) Letting f(w k ) = f (w k1, w k2, …, w km ) be a function over R m, then Defining

18 President UniversityErwin SitompulNNFL 2/18 ff ww ff ww df : positive df : zero df : negative go uphill plain go downhill ff ww To minimize f, we choose Single Layer PerceptronsNeural Networks Gradient Operator df is thus guaranteed to be always negative

19 President UniversityErwin SitompulNNFL 2/19 Single Layer PerceptronsNeural Networks Adaline Learning Rule With then As already obtained before, Weight Modification Rule Defining we can write

20 President UniversityErwin SitompulNNFL 2/20 Single Layer PerceptronsNeural Networks Adaline Learning Modes Batch Learning Mode Incremental Learning Mode

21 President UniversityErwin SitompulNNFL 2/21   - Learning Rule  LMS Algorithm  Widrow-Hoff Learning Rule Single Layer PerceptronsNeural Networks Adaline Learning Rule

22 President UniversityErwin SitompulNNFL 2/22 Single Layer PerceptronsNeural Networks Generalization and Early Stopping By proper training, a neural network may produce reasonable output for inputs not seen during training  Generalization Generalization is particularly useful for the analysis of a “noisy” data (e.g. time–series ) “Overtraining” will not improve the ability of a neural network to produce good output. On the contrary, it will try to take noise as the real data and lost its generality.

23 President UniversityErwin SitompulNNFL 2/23 Generalization and Early Stopping Single Layer PerceptronsNeural Networks Overfitting vs Generalization

24 President UniversityErwin SitompulNNFL 2/24 Homework 2 Single Layer PerceptronsNeural Networks Given a function y = 4x 2, you are required to find the value of x that will result y = 2 by using the Least Mean Squares method. Use initial estimate x 0 = 1 and learning rate η = 0.01. Write down the results of the first 10 epochs/iterations. Give conclusion about your result. Note: Calculation can be done manually or using Matlab.

25 President UniversityErwin SitompulNNFL 2/25 Homework 2A Single Layer PerceptronsNeural Networks Given a function y = 2x 3 + cos 2 x, you are required to find the value of x that will result y = 5 by using the Least Mean Squares method. Use initial estimate x 0 = 0.2*Student ID and learning rate η = 0.01. Write down the results of the first 10 epochs/iterations. Give conclusion about your result. Note: Calculation can be done manually or using Matlab/Excel.


Download ppt "Lecture 2 Introduction to Neural Networks and Fuzzy Logic President UniversityErwin SitompulNNFL 2/1 Dr.-Ing. Erwin Sitompul President University"

Similar presentations


Ads by Google