Study of a neural network-based system for stability augmentation of an airplane

Size: px
Start display at page:

Download "Study of a neural network-based system for stability augmentation of an airplane"

Transcription

1 Study of a neural network-based system for stability augmentation of an airplane Author: Roger Isanta Navarro Annex 1 Introduction to Neural Networks and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) Supervisors: Oriol Lizandra Dalmases Fatiha Nejjari Akhi-Elarab Aeronautical Engineering September 2013

2 This page intentionally left blank

3 Contents 1 Introduction to neural networks Main applications of Neural Networks Advantages and disadvantages of Neural Networks Biological foundations of Neural Networks The artificial neuron Neural Network typologies Required mathematical and numerical tools Matrix inversion formula Least-Squares Method Recursive Least-Squares Method Method of Steepest Descent ANFIS Networks Fuzzy logic, fuzzy inference systems and the Sugeno fuzzy model Advantages of ANFIS over Multilayer Perceptrons (MLP) Architecture ANFIS Hybrid Learning Forward pass Backward pass i

4 List of figures Figure 1.1 Main neural elements... 3 Figure 1.2 Artificial neuron model... 4 Figure 1.3 Examples of step and sigmoidal activation functions... 5 Figure 2.1 Generic convergence diagram of total output error with constant step Figure 2.2 Generic convergence diagram of total output error with variable step Figure 3.1 The Sugeno fuzzy model Figure 3.2 Diagram representation of a 3-inputs 3-rules ANFIS network ii

5 1 Introduction to neural networks Artificial Neural Networks are nonlinear mapping systems whose structure is based in the observed human and animal nervous systems; they are mathematical approximations to the human brain function. Nevertheless, they are not directly comparable to the brain, nor are their operation principles, for they do not base themselves solely in biological networks operation, they only emulate a very simple portion of human its functions and are not able to simulate the highly complex relational neurobiological processes that occur within it. Processing units are named neurons. Artificial Neural Networks (from now on Neural Networks) comprise a large number of these processing units, each of which receives weighted input from other units (or nodes) and generates a scalar output depending on the available local information stored internally and the information arriving through the incoming weighted connections. A Neural Network is characterized by the following aspects: - A set of processing units or neurons. - An activation state for each unit, equivalent to the output of the unit. - Connections between units, often defined by a weight that modifies the effect of the input signal in the unit. - A propagation rule, which specifies the effective entry of a unit as a function of the external entries. - An activation function based on the effective entry and the previous activation. - An external input corresponding to a term known as bias for each unit. - A method to gather information, corresponding to the learning rule. - An environment in which the system is going to operate. 1.1 Main applications of Neural Networks Applications of Neural Networks are diverse and so are the areas in which they are of use. The main applications of Neural Networks are those in which the inferring approximation and use of functions is required via observation or measure, these tasks classify in following categories: - Function approximation - Data classification - Data processing - System control 1

6 System control is the most recent application of Neural Networks and is the field in which most efforts are invested actually. It is also the field in which is developed this study. Specific areas of application of the Neural Networks are also varied, the most significant follow: - Automotive and transport: automatic pilot systems, failure detection by external vibration detection, truck brakes diagnosis, fleet tracking systems. - Banking and finances: checks and other documents reading, credit applications and risks evaluation, real properties appraisal, loans assessment, price evolution forecasting, fake identifications, signature interpretation and identification, use of credit line analysis. - Electronics: code sequence prediction, elements distribution in integrated circuits, process control, failure analysis, artificial vision, voice recognition. - Manufacturing: production and process control, product analysis and design, failure diagnosis, automatic visual quality inspection. - Medicine: EEG and ECG analysis, prosthesis design, optimization of transplant times, recognition and prediction of infarctions via ECG, reduction of hospital costs. - Robotics: dynamic control of trajectory, controllers, optical systems. - Security: adaptive security codes, cryptography, digital recognition of fingerprints. - Telecommunications: data compression, automation of information services, real time translation of spoken language. - Voice: voice recognition, voice comprehension, vowel classification, transformation from text to voice. 1.2 Advantages and disadvantages of Neural Networks Neural Networks present several advantages over other processing systems, the most significant being: - Neural Networks can synthesize algorithms through a learning process. - To use neural technology it is not necessary to now the mathematical details. It is only required familiarity with the job data. - The solution of nonlinear problems is one of the strengths of the Neural Networks. - Neural Networks are robust. They might fail in some processing elements, but the network keeps on working, as opposed as in traditional programming. 2

7 However, some disadvantages are also characteristic of Neural Networks: - Neural Networks must be trained for each problem. Moreover, multiple tests must be conducted to define the adequate architecture. Training might be long and CPU time consuming. - The training requirement involves large volumes of data. Neural Networks present a complex issue for external observers who would want to make modifications in it. To add new knowledge is necessary to change the interactions of many units so that its unified effect might synthesize this new knowledge. 1.3 Biological foundations of Neural Networks A biological neuron is a cell specialized in information processing. It divides, from a simplified point of view, into the cell body (soma) and two different kinds of ramifications: dendrites and the axon; the firsts drive pulse inputs to the soma and the later transmits the output signals generated by the body of the cell. Figure 1.1 Main neural elements The signals that reach the neuron through the dendrites are weighted by a parameter called weight, associated to the corresponding synapsis. These weights might excite the neuron (positive weight synapsis) or inhibit it (negative weight synapsis). The soma then integrates and combines the different weighted input signals, and emits an output signal depending on the sum of the weighted entries: if the sum is higher or equal to the activation threshold of the neuron, the output is generated and sent through the axon. 3

8 The ability to adjust the signals, by modifying the weight values, results in a learning mechanism which is the training methodology of the artificial neural networks. 1.4 The artificial neuron As stated, an artificial neuron is a mathematical approximation to the biological neuron function. Within the generic artificial neuron two processes occur: the first one is the algebraic sum of the inputs of synapsis and the second is the evaluation of a nonlinear function resulting in its output value. Figure 1.2 Artificial neuron model The algebraic sum that will be lately evaluated by the nonlinear function can be written as a function of the synaptic weights, entry signals and a bias. ( 1.1) where counts the total of incoming connections. The polarization or threshold of the neuron is an external parameter, however it can be considered as additional entry : ( 1.2) is called the activation potential. The neuron output, results from the evaluation of the activation function, which takes the activation potential as argument and may also take the previous output. ( ) ( 1.3) 4

9 If the previous output is not considered as argument, the resulting value of the whole process of integration and evaluation within the neuron can be written as ( ) ( 1.4) Many activation functions are used as step, linear, hyperbolic tangent or sigmoidal functions among others y u y u Figure 1.3 Examples of step and sigmoidal activation functions 1.5 Neural Network typologies Artificial Neural Networks may be classified differentiating between neuron function, single or multilayered networks, training methodologies, whether the flow of information is recurrent or feed-forward, etc. McCulloch and Pits, in 1943, introduced the simple artificial neuron. These are neurons which accept values of zero or one as incoming signals and present an activation threshold of one or two, thus allowing the implementation of a logic OR (in case of one as activation threshold) or a logic AND (in case of the activation threshold being equal to two). The integration of postsynaptic pulses is lineal in this kind of neurons. In more realistic neuron designs the entry signals may be a real value and are subject to varying weights. Different neuron models have been designed to fulfill specific requirements such the use of integration methods other than simple algebraic addition in which the entry signals are defined as functions of time which describe changes in time on the voltage. Other neurons such as the ones used within the ANFIS network proposed in this study have special functions such as the product of all their entries, a quotient or the computation of a function linear or nonlinear to some defined or varying parameters. 5

10 Neural networks may also be differentiated after their number of layers. Simplest layers have only one or two layers, the second usually being a sum of the previous node outputs, an example of such neural networks are Simple Perceptrons, whose goal is usually the division of an n-dimensional space into two sub spaces according to a criterion. A neural network differentiating between whether a color is dark or light may be achieved through the implementation of a Simple Single Layered Perceptron. The frontier of decision, which is equivalent to the reaching of the activation threshold in the neuron, may be defined by one of many activation functions of which some have been mentioned earlier. To solve more complicated problems multilayered networks are usually required. There is a great variety of such networks for they allow considerable specialization on the problem at hand. The two most common multilayered neural networks being the Multilayered Perceptron, an extension of the Simple Perceptron to a multilayer scheme, and the Adaptive Neuro-Fuzzy Inference System (ANFIS) network, which is the one used within this study. The structure of the network and the subsequent flow of information are also significant differences between networks, thus the so-called Feed-Forward networks are those in which the information flows from a layer to a strictly superior layer for all layers and nodes of the network. On the contrary, a recurrent network will be that in which one or more nodes receive as input the output of another node in a subsequent layer. Training methodologies are also a significant distinguishing characteristic and are intrinsically related to the network structure. The linearity or nonlinearity to the global network output of some of the its varying parameters will allow or prevent the use of optimization methods like the Least Squares Method presented in subsections 2.2 and 2.3 and the nonlinearity or a recurrent structure will require the use of other optimization procedures such as derivative-based methods as the Steepest Descent presented in subsection 2.4 or derivative-free methods such as Genetic Algorithms. 6

11 2 Required mathematical and numerical tools Within this section the main mathematical and numerical tools that will later be used to develop the neural network training will be presented. 2.1 Matrix inversion formula The matrix inversion formula will be of importance in the following Recursive Least- Squares Method subsection. This formula states that given two nonsingular square matrices and, then ( 2.1) Proof of this statement can be found in [2]. 2.2 Least-Squares Method Least-Squares Method is a standard method to compute the set of parameters that will best approximate the solution to an overdetermined system. To do so, the sum of the squares of the errors is minimized. Consider a linear system of equations consisting in equations and unknown parameters ( 2.2) where is an matrix, is the unknown parameter vector and is the solution vector. It is obvious that if, the matrix will be squared and the exact solution to the system, provided that is nonsingular, can be easily calculated by ( 2.3) If, which is a very common situation in neural network training, as well as in many other applications, an exact solution is not always possible, either because the model is not appropriate enough to describe the system, or because of the existence of noise or error contamination in the data. Equation ( 2.2) should be modified to account for this error: ( 2.4) As defined previously, the main goal of the method is finding an approximate vector which minimizes the sum of squared error. 7

12 ( 2.5) where is the i th row of. The expression in the previous equation may be expanded, derived and equated to as follows: ( 2.6) ( 2.7) At, ( 2.8) If is non-singular, can be solved: ( 2.9) 2.3 Recursive Least-Squares Method One important drawback of the previously disclosed Least-Squares Method resides in the necessity of inverting the matrix: an operation requiring a high computational cost if is not small enough. Moreover, the Least-Squares Method, as presented previously, does not take advantage of the recently computed values, but given additional equations to the system, requires the total recalculation of the method. The following procedure presents a recursive method to account for an additional equation in the system, that is, an extra training pair, taking in consideration the previous values. From now on, the circumflex symbol denoting the approximate solution will be omitted for simplicity. Considering the unknown parameter vector at step ( 2.10) the same vector at step can be written as: ([ ] [ ]) [ ] [ ] ( 2.11) The following and are introduced ( 2.12) 8

13 ([ ] [ ]) ( 2.13) Using the matrix inversion formula presented at the beginning of this section, the computation of can be rewritten as an incremental formula: ( 2.14) Once known, an incremental expression for can easily be found. ( 2.15) can be eliminated from this expression using the Equation ( 2.10) ( 2.16) [ ] ( 2.17) yielding the final incremental expression for : ( 2.18) The sequential calculation of given, using Recursive Least-Squares Method can be summarized in the following two steps: where the initial equals. { ( 2.19) 2.4 Method of Steepest Descent Descent methods main goal is also the minimization of a function defined on an - dimensional input space [ ]. The objective function might not have linear form with respect to, as opposed as considered in the Least-Squares Method (variations of Least-Squares Method exist for nonlinear models, however these will not be considered since will not be used within this study). Also as opposed to the Least- Squares Method, this local minimum is found iteratively due to the complexity and nonlinearity of. Within one iteration, the next values vector, denoted by, is computed by a step from of size in a direction so that 9

14 ( 2.20) ( 2.21) The computation of step is performed through two procedures: determination of direction and determination of the step size. Many different methods exist whose main difference lies in the computation of the first procedure, while the step size is commonly determined by line minimization. However, some methods do not use line minimization, which is the case of the two methods presented below. The taken corresponds to the direction in which decreases more quickly, that is. ( 2.22) Equation ( 2.22) is commonly named simple steepest descent. In it, given a fixed the magnitude of the step varies automatically because of the different gradients of. Experimental results [2] show that the search is not efficient enough for and will not converge for. A good value to choose is close to. Choosing constant presents a considerable oscillation around the optimal solution since there is no real control of the step size, as it depends directly on the value of. This effect is shown in Figure 2.1. Figure 2.1 Generic convergence diagram of total output error with constant step Chang and Fallside introduced a heuristic method known as Backpropagation learning rule with a momentum to update in which a descent direction is influenced by the 10

15 previous one, and the step size increases when the direction looks good according to the relation between both directions [2]. If the gradient vector is normalized, Equation ( 2.23) is named normalized version of steepest descent: ( 2.23) where is the actual Euclidean distance from to. Again, may be chosen fixed or variable. If fixed, a small value will lead to an inefficient search, while a high value will cause an efficient approach to the objective value, but will then oscillate around it, loosing precision. Figure 2.2 Generic convergence diagram of total output error with variable step This automatically forces to consider a dynamically updated value of simple rules: according to two 1. If the objective function undergoes consecutive reductions, increase by. 2. If the objective function undergoes consecutive combinations of one increase one decrease, reduce by. The values of and are typically set to and respectively, after experimental observations. 11

16 3 ANFIS Networks Adaptive Neuro-Fuzzy Inference Systems (ANFIS) are a class of adaptive neural networks that are functionally equivalent to fuzzy inference systems (described in the following subsection) and offer the combination of learning, adaptability and nonlinear, timevariant problem solving characteristics of Artificial Neural Networks plus the important concepts of approximate reasoning and treatment of information provided by the fuzzy set theory. ANFIS network control systems (or neuro-fuzzy systems) represent a hybrid platform for solving actual complex problems that require the use of intelligent systems and are a viable alternative to the conventional model-based control schemes. They allow dealing effectively with the common issues of uncertainty and unknown variations in plant parameters and structure, hence improving robustness of the control system. 3.1 Fuzzy logic, fuzzy inference systems and the Sugeno fuzzy model Fuzzy logic [2] is a set of mathematical principles based on degrees of membership to pre-established functions whose main goal is information modeling; it is a flexible tool based on linguistic rules dictated by an expert. Fuzzy logic was developed to emulate human logic and attain correct solutions in spite of the ambiguity of information. In contrast with conventional logic where strict boundaries are set between the membership of a variable to a set or another, fuzzy logic presents membership ranks within the interval between the two sets, and offers a solution based on this dual or higher membership. As a simple example, the speed of a car may be classified as high or low for a given circumstance, and the reaction of the driver when breaking will depend, among other things, on whether he assigns his current speed to a set or another, or an intermediate state between them. On one hand, conventional logic will establish strict boundaries between the proposed sets; for example, driving at less than will be considered slow and doing so at or more will be considered high speed. Clearly a problem arises when driving at speeds around, since the response of the driver subject to conventional logic will vary abruptly when crossing this boundary. On the other hand, a driver subject to fuzzy logic reasoning will be able to assign partial memberships to both functions, that is, considering partially high and partially slow, and provide a much more accurate response. Moreover, if the parameters defining the membership functions are variable, the fuzzy system will be able, provided a correct training algorithm, to modify such functions to offer a better response. 12

17 The Sugeno fuzzy model aims to generate a systematic approach towards generating fuzzy rules from a given input-output data set. Sugeno fuzzy rules are of the form: ( 3.1) where and are fuzzy sets and is a function associated to the fact of and pertaining to and respectively. will usually be a polynomial, in what is then called the first-order Sugeno fuzzy model, in contrast with the zero-order Sugeno fuzzy model if were constant. Figure 3.1 shows the fuzzy reasoning procedure for a first-order Sugeno fuzzy model: Figure 3.1 The Sugeno fuzzy model 3.2 Advantages of ANFIS over Multilayer Perceptrons (MLP) In addition to the general advantages and disadvantages of the Neural Networks, ANFIS networks present interesting advantages over Multilayer Perceptrons (MLP), which are the most direct competitors in neural computing for the type of problem treated within this study. These advantages result from the fact that ANFIS presents a much more specific mathematical structure which enables it as a good universal adaptive approximator. The most significant advantages of ANFIS in front of MLPs follow: 1. ANFIS presents a much better learning ability: for a similar network complexity, a much smaller convergence error is achieved, and although the convergence is slower the smallness of the error in ANFIS is able to compensate that fact. 2. MLP often present a sudden convergence preceded by a region of considerable instability. 3. ANFIS can achieve highly nonlinear mapping, far superior to MPL and other common linear methods of similar complexity. 13

18 4. ANFIS requires fewer adjustable parameters than those required in other Neural Network structures and, specifically, backpropagation MPLs. 5. The ANFIS structure allows for parallel computation. Finally, ANFIS presents two advantages exclusive to its method: 6. ANFIS networks present a well-structured knowledge representation. 7. ANFIS networks allow a better integration with other control design methods. 3.3 Architecture ANFIS structure consists in layers of neurons, each of which having a very specific behavior. From these, layers, and have a constant behavior, while layers 1 and 4 have varying parameters, the modification of which allows for the network training. Figure 3.2 Diagram representation of a 3-inputs 3-rules ANFIS network The definition of every node within the ANFIS structure follows. All nodes within a same layer have identical behavior, subject to the dependence to varying parameters (layers and ). Considering an -input and -rules ANFIS, the following nodal behaviors are defined: LAYER 1 Layer 1 consists of adaptive neurons in which the fuzzification is performed, that is: the grade of membership to the defined membership functions of the input is evaluated. 14

19 ( 3.2) where input. is the input vector and denotes the neuron in of layer, associated to The membership function may vary; triangular and bell shaped functions are commonly used. A generalized bell-shape function has been chosen for it is one of the most common membership functions and is continuous, which allows better differentiation when performing the backpropagation training: ( ) ( 3.3) This function depends on three parameters: - modifies the bell width. - together with, modifies the bell slope at the point where. - modifies the center position of the bell. These parameters are commonly called premise parameters. LAYER 2 Layer 2 nodes are fixed in which its output is the product of all their entries. The nodes are commonly labeled and their respective outputs represent the firing strength of the rules, or the rules inferences. Depending on the problem at hand there will be a relationship between inputs, and the whole set of possible combinations of each input rule will not be necessarily computed. If however it is necessary to compute all of the possible combinations or a subset of them the definition of combination: will depend on those values of its corresponding ( 3.4) LAYER 3 Layer 3 nodes, labeled are also fixes and their respective outputs represent the normalized inferences, that is: the ratio of the corresponding rule s firing strength to the sum of all the rule s firing strengths. 15

20 ( 3.5) LAYER 4 Nodes in layer 4 are adaptive nodes in which the consequent evaluation inference is calculated, its output is defined as: ( ) ( 3.6) where the set of parameters and is commonly called consequent parameters. LAYER 5 Finally, a single node in layer sums all the outputs from layer to compute the overall output of the network: ( 3.7) 3.4 ANFIS Hybrid Learning ANFIS networks, as any other neural network, might be trained by backpropagation of the resulting error and adjustment of the adaptive parameters according to this propagation in order to minimize it. Nevertheless, ANFIS presents some linearity with respect to some of its parameters, due to its structure, allowing for the application of the much efficient Least Squares Method. The use of the combination of both Least- Squares and Steepest Descent methods is referred to as Hybrid Learning. As commented, the adaptive parameters of the ANFIS network divides into premise parameters in layer 1, and consequent parameters in layer 4. The contribution of these last set to the network output is linear: ( ) ( 3.8) therefore, their computation using Least-Square Method or its recursive version is advised, for the exact is obtained with no need of several iterations and requires less computational effort and time on what a process named forward pass. On the contrary, premise parameters will be computed using the Steepest Descent Method, by 16

21 backpropagating the error through the network, during the backward pass. Both training steps constitute the Hybrid Learning methodology and are presented below Forward pass In the forward pass the Recursive Least-Square Method is used to evaluate the consequent parameters. Considering the notation in subsections 2.2 and 2.3 and a - input network as the one shown in Figure 3.2, the vector will be defined as follows: { } ( 3.9) and will be the desired output vector (not to be confused with the network results, after Layer 5). Considering that the required number of training pairs to have a definite system will be, where it is recalled refers to the number of inputs and to the number of rules, the vector would be as follows: { } ( 3.10) Meanwhile, matrix can be defined after Equations ( 3.6) and ( 3.7): ( 3.11) [ ] where the superscript between parenthesis denotes the training pair number and been defined to simplify the notation. has Within the recursive part of the method, the vector pair is defined as any row of the matrix : corresponding to the training { } ( 3.12) Backward pass In the backward pass the error signal propagates backwards through the network until the dependence of this error to each of the premise parameters is evaluated. Once known this gradient, the parameters may be updated by Steepest Descent: Chain rule is used to evaluate the partial derivatives: ( 3.13) ( 3.14) 17

22 where here stands for any premise parameter, or. The partial derivatives are derived as: [ ] ( 3.15) [ ( )] ( 3.16) { ( ) ( ) ( ) ( 3.17) ( 3.18) The derivative depends on the membership function used, and will be different for each premise parameter, or. To simplify the notation, the following function is introduced: ( 3.19) It is also noticeable, that the absolute value within expression ( 3.3) may be rewritten in order to ease differentiation: [ ] ( 3.20) The following partial derivatives result. For premise parameter : ( 3.21) For : ( 3.22) And for : ( 3.23) 18

23 This page intentionally left blank

Lecture 7 Artificial neural networks: Supervised learning

Lecture 7 Artificial neural networks: Supervised learning Lecture 7 Artificial neural networks: Supervised learning Introduction, or how the brain works The neuron as a simple computing element The perceptron Multilayer neural networks Accelerated learning in

More information

4. Multilayer Perceptrons

4. Multilayer Perceptrons 4. Multilayer Perceptrons This is a supervised error-correction learning algorithm. 1 4.1 Introduction A multilayer feedforward network consists of an input layer, one or more hidden layers, and an output

More information

(Feed-Forward) Neural Networks Dr. Hajira Jabeen, Prof. Jens Lehmann

(Feed-Forward) Neural Networks Dr. Hajira Jabeen, Prof. Jens Lehmann (Feed-Forward) Neural Networks 2016-12-06 Dr. Hajira Jabeen, Prof. Jens Lehmann Outline In the previous lectures we have learned about tensors and factorization methods. RESCAL is a bilinear model for

More information

Artificial Neural Network

Artificial Neural Network Artificial Neural Network Contents 2 What is ANN? Biological Neuron Structure of Neuron Types of Neuron Models of Neuron Analogy with human NN Perceptron OCR Multilayer Neural Network Back propagation

More information

EEE 241: Linear Systems

EEE 241: Linear Systems EEE 4: Linear Systems Summary # 3: Introduction to artificial neural networks DISTRIBUTED REPRESENTATION An ANN consists of simple processing units communicating with each other. The basic elements of

More information

ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD

ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD ARTIFICIAL NEURAL NETWORK PART I HANIEH BORHANAZAD WHAT IS A NEURAL NETWORK? The simplest definition of a neural network, more properly referred to as an 'artificial' neural network (ANN), is provided

More information

Lecture 4: Feed Forward Neural Networks

Lecture 4: Feed Forward Neural Networks Lecture 4: Feed Forward Neural Networks Dr. Roman V Belavkin Middlesex University BIS4435 Biological neurons and the brain A Model of A Single Neuron Neurons as data-driven models Neural Networks Training

More information

Neural Networks and the Back-propagation Algorithm

Neural Networks and the Back-propagation Algorithm Neural Networks and the Back-propagation Algorithm Francisco S. Melo In these notes, we provide a brief overview of the main concepts concerning neural networks and the back-propagation algorithm. We closely

More information

Neural networks. Chapter 20. Chapter 20 1

Neural networks. Chapter 20. Chapter 20 1 Neural networks Chapter 20 Chapter 20 1 Outline Brains Neural networks Perceptrons Multilayer networks Applications of neural networks Chapter 20 2 Brains 10 11 neurons of > 20 types, 10 14 synapses, 1ms

More information

Neural networks. Chapter 19, Sections 1 5 1

Neural networks. Chapter 19, Sections 1 5 1 Neural networks Chapter 19, Sections 1 5 Chapter 19, Sections 1 5 1 Outline Brains Neural networks Perceptrons Multilayer perceptrons Applications of neural networks Chapter 19, Sections 1 5 2 Brains 10

More information

Data Mining Part 5. Prediction

Data Mining Part 5. Prediction Data Mining Part 5. Prediction 5.5. Spring 2010 Instructor: Dr. Masoud Yaghini Outline How the Brain Works Artificial Neural Networks Simple Computing Elements Feed-Forward Networks Perceptrons (Single-layer,

More information

Artifical Neural Networks

Artifical Neural Networks Neural Networks Artifical Neural Networks Neural Networks Biological Neural Networks.................................. Artificial Neural Networks................................... 3 ANN Structure...........................................

More information

EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, Sasidharan Sreedharan

EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, Sasidharan Sreedharan EE04 804(B) Soft Computing Ver. 1.2 Class 2. Neural Networks - I Feb 23, 2012 Sasidharan Sreedharan www.sasidharan.webs.com 3/1/2012 1 Syllabus Artificial Intelligence Systems- Neural Networks, fuzzy logic,

More information

Revision: Neural Network

Revision: Neural Network Revision: Neural Network Exercise 1 Tell whether each of the following statements is true or false by checking the appropriate box. Statement True False a) A perceptron is guaranteed to perfectly learn

More information

Pattern Recognition Prof. P. S. Sastry Department of Electronics and Communication Engineering Indian Institute of Science, Bangalore

Pattern Recognition Prof. P. S. Sastry Department of Electronics and Communication Engineering Indian Institute of Science, Bangalore Pattern Recognition Prof. P. S. Sastry Department of Electronics and Communication Engineering Indian Institute of Science, Bangalore Lecture - 27 Multilayer Feedforward Neural networks with Sigmoidal

More information

AI Programming CS F-20 Neural Networks

AI Programming CS F-20 Neural Networks AI Programming CS662-2008F-20 Neural Networks David Galles Department of Computer Science University of San Francisco 20-0: Symbolic AI Most of this class has been focused on Symbolic AI Focus or symbols

More information

2015 Todd Neller. A.I.M.A. text figures 1995 Prentice Hall. Used by permission. Neural Networks. Todd W. Neller

2015 Todd Neller. A.I.M.A. text figures 1995 Prentice Hall. Used by permission. Neural Networks. Todd W. Neller 2015 Todd Neller. A.I.M.A. text figures 1995 Prentice Hall. Used by permission. Neural Networks Todd W. Neller Machine Learning Learning is such an important part of what we consider "intelligence" that

More information

Neural Networks. Chapter 18, Section 7. TB Artificial Intelligence. Slides from AIMA 1/ 21

Neural Networks. Chapter 18, Section 7. TB Artificial Intelligence. Slides from AIMA   1/ 21 Neural Networks Chapter 8, Section 7 TB Artificial Intelligence Slides from AIMA http://aima.cs.berkeley.edu / 2 Outline Brains Neural networks Perceptrons Multilayer perceptrons Applications of neural

More information

A FUZZY NEURAL NETWORK MODEL FOR FORECASTING STOCK PRICE

A FUZZY NEURAL NETWORK MODEL FOR FORECASTING STOCK PRICE A FUZZY NEURAL NETWORK MODEL FOR FORECASTING STOCK PRICE Li Sheng Institute of intelligent information engineering Zheiang University Hangzhou, 3007, P. R. China ABSTRACT In this paper, a neural network-driven

More information

Neural Networks and Fuzzy Logic Rajendra Dept.of CSE ASCET

Neural Networks and Fuzzy Logic Rajendra Dept.of CSE ASCET Unit-. Definition Neural network is a massively parallel distributed processing system, made of highly inter-connected neural computing elements that have the ability to learn and thereby acquire knowledge

More information

An artificial neural networks (ANNs) model is a functional abstraction of the

An artificial neural networks (ANNs) model is a functional abstraction of the CHAPER 3 3. Introduction An artificial neural networs (ANNs) model is a functional abstraction of the biological neural structures of the central nervous system. hey are composed of many simple and highly

More information

ECE521 Lectures 9 Fully Connected Neural Networks

ECE521 Lectures 9 Fully Connected Neural Networks ECE521 Lectures 9 Fully Connected Neural Networks Outline Multi-class classification Learning multi-layer neural networks 2 Measuring distance in probability space We learnt that the squared L2 distance

More information

Artificial Neural Network and Fuzzy Logic

Artificial Neural Network and Fuzzy Logic Artificial Neural Network and Fuzzy Logic 1 Syllabus 2 Syllabus 3 Books 1. Artificial Neural Networks by B. Yagnanarayan, PHI - (Cover Topologies part of unit 1 and All part of Unit 2) 2. Neural Networks

More information

Last update: October 26, Neural networks. CMSC 421: Section Dana Nau

Last update: October 26, Neural networks. CMSC 421: Section Dana Nau Last update: October 26, 207 Neural networks CMSC 42: Section 8.7 Dana Nau Outline Applications of neural networks Brains Neural network units Perceptrons Multilayer perceptrons 2 Example Applications

More information

Neural Networks and Ensemble Methods for Classification

Neural Networks and Ensemble Methods for Classification Neural Networks and Ensemble Methods for Classification NEURAL NETWORKS 2 Neural Networks A neural network is a set of connected input/output units (neurons) where each connection has a weight associated

More information

ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92

ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92 ARTIFICIAL NEURAL NETWORKS گروه مطالعاتي 17 بهار 92 BIOLOGICAL INSPIRATIONS Some numbers The human brain contains about 10 billion nerve cells (neurons) Each neuron is connected to the others through 10000

More information

CS 4700: Foundations of Artificial Intelligence

CS 4700: Foundations of Artificial Intelligence CS 4700: Foundations of Artificial Intelligence Prof. Bart Selman selman@cs.cornell.edu Machine Learning: Neural Networks R&N 18.7 Intro & perceptron learning 1 2 Neuron: How the brain works # neurons

More information

Lecture 4: Perceptrons and Multilayer Perceptrons

Lecture 4: Perceptrons and Multilayer Perceptrons Lecture 4: Perceptrons and Multilayer Perceptrons Cognitive Systems II - Machine Learning SS 2005 Part I: Basic Approaches of Concept Learning Perceptrons, Artificial Neuronal Networks Lecture 4: Perceptrons

More information

Machine Learning and Data Mining. Multi-layer Perceptrons & Neural Networks: Basics. Prof. Alexander Ihler

Machine Learning and Data Mining. Multi-layer Perceptrons & Neural Networks: Basics. Prof. Alexander Ihler + Machine Learning and Data Mining Multi-layer Perceptrons & Neural Networks: Basics Prof. Alexander Ihler Linear Classifiers (Perceptrons) Linear Classifiers a linear classifier is a mapping which partitions

More information

Artificial Neural Networks" and Nonparametric Methods" CMPSCI 383 Nov 17, 2011!

Artificial Neural Networks and Nonparametric Methods CMPSCI 383 Nov 17, 2011! Artificial Neural Networks" and Nonparametric Methods" CMPSCI 383 Nov 17, 2011! 1 Todayʼs lecture" How the brain works (!)! Artificial neural networks! Perceptrons! Multilayer feed-forward networks! Error

More information

Neural Networks biological neuron artificial neuron 1

Neural Networks biological neuron artificial neuron 1 Neural Networks biological neuron artificial neuron 1 A two-layer neural network Output layer (activation represents classification) Weighted connections Hidden layer ( internal representation ) Input

More information

Neural networks. Chapter 20, Section 5 1

Neural networks. Chapter 20, Section 5 1 Neural networks Chapter 20, Section 5 Chapter 20, Section 5 Outline Brains Neural networks Perceptrons Multilayer perceptrons Applications of neural networks Chapter 20, Section 5 2 Brains 0 neurons of

More information

Chapter 9: The Perceptron

Chapter 9: The Perceptron Chapter 9: The Perceptron 9.1 INTRODUCTION At this point in the book, we have completed all of the exercises that we are going to do with the James program. These exercises have shown that distributed

More information

CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR

CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR CHAPTER 4 FUZZY AND NEURAL NETWORK FOR SR MOTOR 4.1 Introduction Fuzzy Logic control is based on fuzzy set theory. A fuzzy set is a set having uncertain and imprecise nature of abstract thoughts, concepts

More information

ADAPTIVE NEURO-FUZZY INFERENCE SYSTEMS

ADAPTIVE NEURO-FUZZY INFERENCE SYSTEMS ADAPTIVE NEURO-FUZZY INFERENCE SYSTEMS RBFN and TS systems Equivalent if the following hold: Both RBFN and TS use same aggregation method for output (weighted sum or weighted average) Number of basis functions

More information

Introduction Biologically Motivated Crude Model Backpropagation

Introduction Biologically Motivated Crude Model Backpropagation Introduction Biologically Motivated Crude Model Backpropagation 1 McCulloch-Pitts Neurons In 1943 Warren S. McCulloch, a neuroscientist, and Walter Pitts, a logician, published A logical calculus of the

More information

SPSS, University of Texas at Arlington. Topics in Machine Learning-EE 5359 Neural Networks

SPSS, University of Texas at Arlington. Topics in Machine Learning-EE 5359 Neural Networks Topics in Machine Learning-EE 5359 Neural Networks 1 The Perceptron Output: A perceptron is a function that maps D-dimensional vectors to real numbers. For notational convenience, we add a zero-th dimension

More information

Unit 8: Introduction to neural networks. Perceptrons

Unit 8: Introduction to neural networks. Perceptrons Unit 8: Introduction to neural networks. Perceptrons D. Balbontín Noval F. J. Martín Mateos J. L. Ruiz Reina A. Riscos Núñez Departamento de Ciencias de la Computación e Inteligencia Artificial Universidad

More information

ARTIFICIAL INTELLIGENCE. Artificial Neural Networks

ARTIFICIAL INTELLIGENCE. Artificial Neural Networks INFOB2KI 2017-2018 Utrecht University The Netherlands ARTIFICIAL INTELLIGENCE Artificial Neural Networks Lecturer: Silja Renooij These slides are part of the INFOB2KI Course Notes available from www.cs.uu.nl/docs/vakken/b2ki/schema.html

More information

Artificial Neural Networks Examination, June 2005

Artificial Neural Networks Examination, June 2005 Artificial Neural Networks Examination, June 2005 Instructions There are SIXTY questions. (The pass mark is 30 out of 60). For each question, please select a maximum of ONE of the given answers (either

More information

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others)

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others) Machine Learning Neural Networks (slides from Domingos, Pardo, others) Human Brain Neurons Input-Output Transformation Input Spikes Output Spike Spike (= a brief pulse) (Excitatory Post-Synaptic Potential)

More information

Neural Networks. Nicholas Ruozzi University of Texas at Dallas

Neural Networks. Nicholas Ruozzi University of Texas at Dallas Neural Networks Nicholas Ruozzi University of Texas at Dallas Handwritten Digit Recognition Given a collection of handwritten digits and their corresponding labels, we d like to be able to correctly classify

More information

Part 8: Neural Networks

Part 8: Neural Networks METU Informatics Institute Min720 Pattern Classification ith Bio-Medical Applications Part 8: Neural Netors - INTRODUCTION: BIOLOGICAL VS. ARTIFICIAL Biological Neural Netors A Neuron: - A nerve cell as

More information

Feedforward Neural Nets and Backpropagation

Feedforward Neural Nets and Backpropagation Feedforward Neural Nets and Backpropagation Julie Nutini University of British Columbia MLRG September 28 th, 2016 1 / 23 Supervised Learning Roadmap Supervised Learning: Assume that we are given the features

More information

Linear discriminant functions

Linear discriminant functions Andrea Passerini passerini@disi.unitn.it Machine Learning Discriminative learning Discriminative vs generative Generative learning assumes knowledge of the distribution governing the data Discriminative

More information

y(x n, w) t n 2. (1)

y(x n, w) t n 2. (1) Network training: Training a neural network involves determining the weight parameter vector w that minimizes a cost function. Given a training set comprising a set of input vector {x n }, n = 1,...N,

More information

Artificial neural networks

Artificial neural networks Artificial neural networks Chapter 8, Section 7 Artificial Intelligence, spring 203, Peter Ljunglöf; based on AIMA Slides c Stuart Russel and Peter Norvig, 2004 Chapter 8, Section 7 Outline Brains Neural

More information

Analysis of Multilayer Neural Network Modeling and Long Short-Term Memory

Analysis of Multilayer Neural Network Modeling and Long Short-Term Memory Analysis of Multilayer Neural Network Modeling and Long Short-Term Memory Danilo López, Nelson Vera, Luis Pedraza International Science Index, Mathematical and Computational Sciences waset.org/publication/10006216

More information

Neural Networks Learning the network: Backprop , Fall 2018 Lecture 4

Neural Networks Learning the network: Backprop , Fall 2018 Lecture 4 Neural Networks Learning the network: Backprop 11-785, Fall 2018 Lecture 4 1 Recap: The MLP can represent any function The MLP can be constructed to represent anything But how do we construct it? 2 Recap:

More information

Serious limitations of (single-layer) perceptrons: Cannot learn non-linearly separable tasks. Cannot approximate (learn) non-linear functions

Serious limitations of (single-layer) perceptrons: Cannot learn non-linearly separable tasks. Cannot approximate (learn) non-linear functions BACK-PROPAGATION NETWORKS Serious limitations of (single-layer) perceptrons: Cannot learn non-linearly separable tasks Cannot approximate (learn) non-linear functions Difficult (if not impossible) to design

More information

CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska. NEURAL NETWORKS Learning

CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska. NEURAL NETWORKS Learning CSE 352 (AI) LECTURE NOTES Professor Anita Wasilewska NEURAL NETWORKS Learning Neural Networks Classifier Short Presentation INPUT: classification data, i.e. it contains an classification (class) attribute.

More information

Artificial Neural Networks. Historical description

Artificial Neural Networks. Historical description Artificial Neural Networks Historical description Victor G. Lopez 1 / 23 Artificial Neural Networks (ANN) An artificial neural network is a computational model that attempts to emulate the functions of

More information

CMSC 421: Neural Computation. Applications of Neural Networks

CMSC 421: Neural Computation. Applications of Neural Networks CMSC 42: Neural Computation definition synonyms neural networks artificial neural networks neural modeling connectionist models parallel distributed processing AI perspective Applications of Neural Networks

More information

Unit III. A Survey of Neural Network Model

Unit III. A Survey of Neural Network Model Unit III A Survey of Neural Network Model 1 Single Layer Perceptron Perceptron the first adaptive network architecture was invented by Frank Rosenblatt in 1957. It can be used for the classification of

More information

8. Lecture Neural Networks

8. Lecture Neural Networks Soft Control (AT 3, RMA) 8. Lecture Neural Networks Learning Process Contents of the 8 th lecture 1. Introduction of Soft Control: Definition and Limitations, Basics of Intelligent" Systems 2. Knowledge

More information

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others)

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others) Machine Learning Neural Networks (slides from Domingos, Pardo, others) For this week, Reading Chapter 4: Neural Networks (Mitchell, 1997) See Canvas For subsequent weeks: Scaling Learning Algorithms toward

More information

Artificial Neural Networks The Introduction

Artificial Neural Networks The Introduction Artificial Neural Networks The Introduction 01001110 01100101 01110101 01110010 01101111 01101110 01101111 01110110 01100001 00100000 01110011 01101011 01110101 01110000 01101001 01101110 01100001 00100000

More information

Neural Networks Introduction

Neural Networks Introduction Neural Networks Introduction H.A Talebi Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Winter 2011 H. A. Talebi, Farzaneh Abdollahi Neural Networks 1/22 Biological

More information

Instituto Tecnológico y de Estudios Superiores de Occidente Departamento de Electrónica, Sistemas e Informática. Introductory Notes on Neural Networks

Instituto Tecnológico y de Estudios Superiores de Occidente Departamento de Electrónica, Sistemas e Informática. Introductory Notes on Neural Networks Introductory Notes on Neural Networs Dr. José Ernesto Rayas Sánche April Introductory Notes on Neural Networs Dr. José Ernesto Rayas Sánche BIOLOGICAL NEURAL NETWORKS The brain can be seen as a highly

More information

CHAPTER 4 BASICS OF ULTRASONIC MEASUREMENT AND ANFIS MODELLING

CHAPTER 4 BASICS OF ULTRASONIC MEASUREMENT AND ANFIS MODELLING 37 CHAPTER 4 BASICS OF ULTRASONIC MEASUREMENT AND ANFIS MODELLING 4.1 BASICS OF ULTRASONIC MEASUREMENT All sound waves, whether audible or ultrasonic, are mechanical vibrations involving movement in the

More information

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others)

Machine Learning. Neural Networks. (slides from Domingos, Pardo, others) Machine Learning Neural Networks (slides from Domingos, Pardo, others) For this week, Reading Chapter 4: Neural Networks (Mitchell, 1997) See Canvas For subsequent weeks: Scaling Learning Algorithms toward

More information

Introduction to Machine Learning Spring 2018 Note Neural Networks

Introduction to Machine Learning Spring 2018 Note Neural Networks CS 189 Introduction to Machine Learning Spring 2018 Note 14 1 Neural Networks Neural networks are a class of compositional function approximators. They come in a variety of shapes and sizes. In this class,

More information

Multilayer Neural Networks. (sometimes called Multilayer Perceptrons or MLPs)

Multilayer Neural Networks. (sometimes called Multilayer Perceptrons or MLPs) Multilayer Neural Networks (sometimes called Multilayer Perceptrons or MLPs) Linear separability Hyperplane In 2D: w x + w 2 x 2 + w 0 = 0 Feature x 2 = w w 2 x w 0 w 2 Feature 2 A perceptron can separate

More information

Machine Learning. Neural Networks

Machine Learning. Neural Networks Machine Learning Neural Networks Bryan Pardo, Northwestern University, Machine Learning EECS 349 Fall 2007 Biological Analogy Bryan Pardo, Northwestern University, Machine Learning EECS 349 Fall 2007 THE

More information

Machine Learning for Large-Scale Data Analysis and Decision Making A. Neural Networks Week #6

Machine Learning for Large-Scale Data Analysis and Decision Making A. Neural Networks Week #6 Machine Learning for Large-Scale Data Analysis and Decision Making 80-629-17A Neural Networks Week #6 Today Neural Networks A. Modeling B. Fitting C. Deep neural networks Today s material is (adapted)

More information

Learning and Memory in Neural Networks

Learning and Memory in Neural Networks Learning and Memory in Neural Networks Guy Billings, Neuroinformatics Doctoral Training Centre, The School of Informatics, The University of Edinburgh, UK. Neural networks consist of computational units

More information

Artificial Neural Networks

Artificial Neural Networks Introduction ANN in Action Final Observations Application: Poverty Detection Artificial Neural Networks Alvaro J. Riascos Villegas University of los Andes and Quantil July 6 2018 Artificial Neural Networks

More information

Introduction to Artificial Neural Networks

Introduction to Artificial Neural Networks Facultés Universitaires Notre-Dame de la Paix 27 March 2007 Outline 1 Introduction 2 Fundamentals Biological neuron Artificial neuron Artificial Neural Network Outline 3 Single-layer ANN Perceptron Adaline

More information

Machine Learning. Neural Networks. Le Song. CSE6740/CS7641/ISYE6740, Fall Lecture 7, September 11, 2012 Based on slides from Eric Xing, CMU

Machine Learning. Neural Networks. Le Song. CSE6740/CS7641/ISYE6740, Fall Lecture 7, September 11, 2012 Based on slides from Eric Xing, CMU Machine Learning CSE6740/CS7641/ISYE6740, Fall 2012 Neural Networks Le Song Lecture 7, September 11, 2012 Based on slides from Eric Xing, CMU Reading: Chap. 5 CB Learning highly non-linear functions f:

More information

Multilayer Perceptrons (MLPs)

Multilayer Perceptrons (MLPs) CSE 5526: Introduction to Neural Networks Multilayer Perceptrons (MLPs) 1 Motivation Multilayer networks are more powerful than singlelayer nets Example: XOR problem x 2 1 AND x o x 1 x 2 +1-1 o x x 1-1

More information

Introduction to Machine Learning

Introduction to Machine Learning Introduction to Machine Learning Neural Networks Varun Chandola x x 5 Input Outline Contents February 2, 207 Extending Perceptrons 2 Multi Layered Perceptrons 2 2. Generalizing to Multiple Labels.................

More information

ECE521 Lecture 7/8. Logistic Regression

ECE521 Lecture 7/8. Logistic Regression ECE521 Lecture 7/8 Logistic Regression Outline Logistic regression (Continue) A single neuron Learning neural networks Multi-class classification 2 Logistic regression The output of a logistic regression

More information

Artificial Intelligence

Artificial Intelligence Artificial Intelligence Jeff Clune Assistant Professor Evolving Artificial Intelligence Laboratory Announcements Be making progress on your projects! Three Types of Learning Unsupervised Supervised Reinforcement

More information

Neural Networks. CSE 6363 Machine Learning Vassilis Athitsos Computer Science and Engineering Department University of Texas at Arlington

Neural Networks. CSE 6363 Machine Learning Vassilis Athitsos Computer Science and Engineering Department University of Texas at Arlington Neural Networks CSE 6363 Machine Learning Vassilis Athitsos Computer Science and Engineering Department University of Texas at Arlington 1 Perceptrons x 0 = 1 x 1 x 2 z = h w T x Output: z x D A perceptron

More information

N. Sarikaya Department of Aircraft Electrical and Electronics Civil Aviation School Erciyes University Kayseri 38039, Turkey

N. Sarikaya Department of Aircraft Electrical and Electronics Civil Aviation School Erciyes University Kayseri 38039, Turkey Progress In Electromagnetics Research B, Vol. 6, 225 237, 2008 ADAPTIVE NEURO-FUZZY INFERENCE SYSTEM FOR THE COMPUTATION OF THE CHARACTERISTIC IMPEDANCE AND THE EFFECTIVE PERMITTIVITY OF THE MICRO-COPLANAR

More information

Artificial Neural Networks

Artificial Neural Networks Artificial Neural Networks Oliver Schulte - CMPT 310 Neural Networks Neural networks arise from attempts to model human/animal brains Many models, many claims of biological plausibility We will focus on

More information

Introduction to Natural Computation. Lecture 9. Multilayer Perceptrons and Backpropagation. Peter Lewis

Introduction to Natural Computation. Lecture 9. Multilayer Perceptrons and Backpropagation. Peter Lewis Introduction to Natural Computation Lecture 9 Multilayer Perceptrons and Backpropagation Peter Lewis 1 / 25 Overview of the Lecture Why multilayer perceptrons? Some applications of multilayer perceptrons.

More information

Multilayer Perceptron Tutorial

Multilayer Perceptron Tutorial Multilayer Perceptron Tutorial Leonardo Noriega School of Computing Staffordshire University Beaconside Staffordshire ST18 0DG email: l.a.noriega@staffs.ac.uk November 17, 2005 1 Introduction to Neural

More information

CS:4420 Artificial Intelligence

CS:4420 Artificial Intelligence CS:4420 Artificial Intelligence Spring 2018 Neural Networks Cesare Tinelli The University of Iowa Copyright 2004 18, Cesare Tinelli and Stuart Russell a a These notes were originally developed by Stuart

More information

POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH

POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH Abstract POWER SYSTEM DYNAMIC SECURITY ASSESSMENT CLASSICAL TO MODERN APPROACH A.H.M.A.Rahim S.K.Chakravarthy Department of Electrical Engineering K.F. University of Petroleum and Minerals Dhahran. Dynamic

More information

Multilayer Neural Networks. (sometimes called Multilayer Perceptrons or MLPs)

Multilayer Neural Networks. (sometimes called Multilayer Perceptrons or MLPs) Multilayer Neural Networks (sometimes called Multilayer Perceptrons or MLPs) Linear separability Hyperplane In 2D: w 1 x 1 + w 2 x 2 + w 0 = 0 Feature 1 x 2 = w 1 w 2 x 1 w 0 w 2 Feature 2 A perceptron

More information

CSC Neural Networks. Perceptron Learning Rule

CSC Neural Networks. Perceptron Learning Rule CSC 302 1.5 Neural Networks Perceptron Learning Rule 1 Objectives Determining the weight matrix and bias for perceptron networks with many inputs. Explaining what a learning rule is. Developing the perceptron

More information

CSC 411 Lecture 10: Neural Networks

CSC 411 Lecture 10: Neural Networks CSC 411 Lecture 10: Neural Networks Roger Grosse, Amir-massoud Farahmand, and Juan Carrasquilla University of Toronto UofT CSC 411: 10-Neural Networks 1 / 35 Inspiration: The Brain Our brain has 10 11

More information

Νεςπο-Ασαυήρ Υπολογιστική Neuro-Fuzzy Computing

Νεςπο-Ασαυήρ Υπολογιστική Neuro-Fuzzy Computing Νεςπο-Ασαυήρ Υπολογιστική Neuro-Fuzzy Computing ΗΥ418 Διδάσκων Δημήτριος Κατσαρός @ Τμ. ΗΜΜΥ Πανεπιστήμιο Θεσσαλίαρ Διάλεξη 4η 1 Perceptron s convergence 2 Proof of convergence Suppose that we have n training

More information

Lecture 6. Regression

Lecture 6. Regression Lecture 6. Regression Prof. Alan Yuille Summer 2014 Outline 1. Introduction to Regression 2. Binary Regression 3. Linear Regression; Polynomial Regression 4. Non-linear Regression; Multilayer Perceptron

More information

Artificial Neural Network Method of Rock Mass Blastability Classification

Artificial Neural Network Method of Rock Mass Blastability Classification Artificial Neural Network Method of Rock Mass Blastability Classification Jiang Han, Xu Weiya, Xie Shouyi Research Institute of Geotechnical Engineering, Hohai University, Nanjing, Jiangshu, P.R.China

More information

18.6 Regression and Classification with Linear Models

18.6 Regression and Classification with Linear Models 18.6 Regression and Classification with Linear Models 352 The hypothesis space of linear functions of continuous-valued inputs has been used for hundreds of years A univariate linear function (a straight

More information

22c145-Fall 01: Neural Networks. Neural Networks. Readings: Chapter 19 of Russell & Norvig. Cesare Tinelli 1

22c145-Fall 01: Neural Networks. Neural Networks. Readings: Chapter 19 of Russell & Norvig. Cesare Tinelli 1 Neural Networks Readings: Chapter 19 of Russell & Norvig. Cesare Tinelli 1 Brains as Computational Devices Brains advantages with respect to digital computers: Massively parallel Fault-tolerant Reliable

More information

CS 4700: Foundations of Artificial Intelligence

CS 4700: Foundations of Artificial Intelligence CS 4700: Foundations of Artificial Intelligence Prof. Bart Selman selman@cs.cornell.edu Machine Learning: Neural Networks R&N 18.7 Intro & perceptron learning 1 2 Neuron: How the brain works # neurons

More information

Neural Networks Lecture 4: Radial Bases Function Networks

Neural Networks Lecture 4: Radial Bases Function Networks Neural Networks Lecture 4: Radial Bases Function Networks H.A Talebi Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Winter 2011. A. Talebi, Farzaneh Abdollahi

More information

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption

Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption Application of Artificial Neural Networks in Evaluation and Identification of Electrical Loss in Transformers According to the Energy Consumption ANDRÉ NUNES DE SOUZA, JOSÉ ALFREDO C. ULSON, IVAN NUNES

More information

MODELLING OF TOOL LIFE, TORQUE AND THRUST FORCE IN DRILLING: A NEURO-FUZZY APPROACH

MODELLING OF TOOL LIFE, TORQUE AND THRUST FORCE IN DRILLING: A NEURO-FUZZY APPROACH ISSN 1726-4529 Int j simul model 9 (2010) 2, 74-85 Original scientific paper MODELLING OF TOOL LIFE, TORQUE AND THRUST FORCE IN DRILLING: A NEURO-FUZZY APPROACH Roy, S. S. Department of Mechanical Engineering,

More information

Neural Networks DWML, /25

Neural Networks DWML, /25 DWML, 2007 /25 Neural networks: Biological and artificial Consider humans: Neuron switching time 0.00 second Number of neurons 0 0 Connections per neuron 0 4-0 5 Scene recognition time 0. sec 00 inference

More information

Feed-forward Networks Network Training Error Backpropagation Applications. Neural Networks. Oliver Schulte - CMPT 726. Bishop PRML Ch.

Feed-forward Networks Network Training Error Backpropagation Applications. Neural Networks. Oliver Schulte - CMPT 726. Bishop PRML Ch. Neural Networks Oliver Schulte - CMPT 726 Bishop PRML Ch. 5 Neural Networks Neural networks arise from attempts to model human/animal brains Many models, many claims of biological plausibility We will

More information

Neural Networks. Xiaojin Zhu Computer Sciences Department University of Wisconsin, Madison. slide 1

Neural Networks. Xiaojin Zhu Computer Sciences Department University of Wisconsin, Madison. slide 1 Neural Networks Xiaoin Zhu erryzhu@cs.wisc.edu Computer Sciences Department University of Wisconsin, Madison slide 1 Terminator 2 (1991) JOHN: Can you learn? So you can be... you know. More human. Not

More information

Introduction To Artificial Neural Networks

Introduction To Artificial Neural Networks Introduction To Artificial Neural Networks Machine Learning Supervised circle square circle square Unsupervised group these into two categories Supervised Machine Learning Supervised Machine Learning Supervised

More information

Simple neuron model Components of simple neuron

Simple neuron model Components of simple neuron Outline 1. Simple neuron model 2. Components of artificial neural networks 3. Common activation functions 4. MATLAB representation of neural network. Single neuron model Simple neuron model Components

More information

Computational Intelligence Lecture 3: Simple Neural Networks for Pattern Classification

Computational Intelligence Lecture 3: Simple Neural Networks for Pattern Classification Computational Intelligence Lecture 3: Simple Neural Networks for Pattern Classification Farzaneh Abdollahi Department of Electrical Engineering Amirkabir University of Technology Fall 2011 arzaneh Abdollahi

More information

A summary of Deep Learning without Poor Local Minima

A summary of Deep Learning without Poor Local Minima A summary of Deep Learning without Poor Local Minima by Kenji Kawaguchi MIT oral presentation at NIPS 2016 Learning Supervised (or Predictive) learning Learn a mapping from inputs x to outputs y, given

More information

Linear Regression, Neural Networks, etc.

Linear Regression, Neural Networks, etc. Linear Regression, Neural Networks, etc. Gradient Descent Many machine learning problems can be cast as optimization problems Define a function that corresponds to learning error. (More on this later)

More information