Artificial Neural Networks

Size: px
Start display at page:

Download "Artificial Neural Networks"

Transcription

1 Artificial Neural Networks Short introduction Bojana Dalbelo Bašić, Marko Čupić, Jan Šnajder Faculty of Electrical Engineering and Computing University of Zagreb Zagreb, June 6, 2018 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

2 Motivation Automated data processing is nowadays mostly carried out by digital computers. However, there is still far more data that are not automatically processed. This data is processed by nervous systems of living organisms! The development of one computer science branch has been motivated by the dominant way of data processing in the world we live in. We are looking for a different concept of data processing which would be more similar to the way the biological brain functions. An AI-system that successfully mimics the functioning of the brain would be intelligent. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

3 Contents 1 Introduction to neuro-computing 2 Artificial neuron 3 Artificial neural network Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

4 Contents 1 Introduction to neuro-computing 2 Artificial neuron 3 Artificial neural network Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

5 Motivation for the development of neuro-computing It is known that the brain consists of a large number of neurons operating in parallel. We also know the following about the human brain: There is more than 100 different types of neurons Each type of neuron performs a very simple function The processing speed of a neuron is: 2 milliseconds per input The number of neurons in the human brain: On average, each neuron acquires inputs from between 10 3 and 10 4 other neurons Information is processed both serially and in parallel Information is analogous The processing is fault-tolerant Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

6 Motivation for the development of neuro-computing We wish to build a computer system that would process data in the same way! New paradigm: artificial neural networks The research area dealing with this aspect of data processing: neuro-computing A branch of computing from the group of soft computing Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

7 Development directions of artificial intelligence Since the first days of AI (early 50s) there are two approaches to developing intelligent systems: The symbolic approach: domain knowledge is represented by a set of atomic semantic objects (symbols) which are then manipulated using algorithmic rules The connectivist approach: based on building an architecture similar to that of the brain which, instead of being programmed, learns based on experience The symbolic approach performs well in many areas (it became especially popular with the development of expert systems), but it did not fulfill very high early expectations. The cause of the failure lies in the assumption that all knowledge can be formalized and that the brain is a machine that processes data using formal rules. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

8 The connectivist approach Many everyday tasks are too complex for symbolic representations, e.g. pattern recognition Our mother we can recognize within 0.1 s Neurons in the brain fire roughly once every millisecond It follows that in the the given time at most 100 neurons can fire Obviously processing is parallel! Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

9 Artificial neural network - definition In a broader sense: an artificial replica of the human brain which seeks to simulate the process of learning and data processing. Artificial neural network An artificial neural network is a set of interlinked simple processing elements (neurons) whose functionality is based on the biological neuron and serves to enable distributed parallel data processing. Enables robust data processing. Can be used for both classification and regression tasks. Capable of learning from data. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

10 Learning an artificial neural network Two phases of using an ANN: 1 The learning (training) phase and 2 The data processing phase (exploatation). Learning is an iterative procedure of presenting the network with input examples (experience) and possibly expected outputs which adapts the weights between the neurons. One iteration of presenting all training examples to the network is called an epoch Different variants of learning: 1 On-line learning: weights are adapted after every train example 2 Minibatch learning: weights are adapted after several training examples 3 Batch learning: weights are adapted only after all training examples have been presented to the network The knowledge of how to convert inputs to desired outputs is represented implicitly in the neuron weights Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

11 Learning an artificial neural network Learning artificial neural networks can be: 1 supervised (learning with a teacher) 2 unsupervised (learning without a teacher) 3 reinforcement learning The set of available training examples is often divided into: 1 Training set: e.g. 70% of examples, serves for iterative adaptation of weights 2 Validation set: e.g. 15% of examples, serves to check the generalization power of the network 3 Test set: e.g. 15% of examples, used for final check of the network performance and comparison with different models Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

12 Learning an artificial neural network Learning is carried out on the training set by minimizing a measure of output errors. Accuracy is also monitored on the validation set (but this set is not used for adapting the weights). Overfitting: the network loses the desirable generalization property and becomes an expert for the train set. To prevent overfitting, we stop learning when the validation set error starts to rise. pogreška prekinuti učenje pogreška na skupu za provjeru pogreška na skupu za učenje epoha Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

13 Contents 1 Introduction to neuro-computing 2 Artificial neuron 3 Artificial neural network Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

14 The biological neuron The biological neuron consists of the body (soma), dendrites and axon In the human brain each neuron is connected to 1000 do other neurons on average Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

15 Artificial neuron McCulloch-Pitts defined a simple model of a biological neuron (1943.): TLU-perceptron (Threshold Logic Unit) The numerical value of each input x i is multiplied by the sensitivity to that input w i and accumulated in the body. To the total sum a bias of w 0 (sometimes denoted b) is also added. ( n This defines the accumulated value as net = ) x i w i + w 0. i=1 This accumulated value is the input of the activation function which produces the neuron output: o = step(net). w 0 x 1 w 1 x 2 w 2 net prijenosna funkcija step(net) o w n x n Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

16 Artificial neuron In general the the artificial neuron model enables the accumulated value to pass through an activation function f. w 0 x 1 w 1 x 2 w 2 net prijenosna funkcija f(net) o w n x n Activation functions often used in practice: Identity (ADALINE-neuron) Step function (TLU-perceptron) Sigmoid (sigmoid neuron) Hyperbolic tangent Hinge (Rectified Linear Unit, ReLU) Leaky hinge (Leaky Rectified Linear Unit, LReLU) Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

17 Activation functions f(net) f(net) net (a) Identity function net (b) Step function Identity function f (net) = net Step function f (net) = step(net) = { 0 net < 0 1 otherwise Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

18 Activation functions f(net) 0 f(net) (c) Sigmoid function net (d) Hyperbolic tangent net Sigmoid function f (net) = sigm(net) = 1 1+e net Hyperbolic tangent f (net) = tanh(net) = 2 1+e 2 net 1 = 2 sigm(2x) 1 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

19 Activation functions f(net) 2 f(net) net (e) Hinge net (f) Leaky hinge with α = 0.05 Hinge f (net) = max(0, net) Leaky hinge { net net >= 0 f (net) = α net otherwise Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

20 Classification Classification is the procedure of assigning labels to examples based on a set of features (e.g., color, shape, weight,...). We wish to build a classification system that, based on examples from the training set, can correctly classify new examples If there are only two classes, we are dealing with binary classification. We can use a system that will have a single output defining two clearly separate states (e.g.: 0 and 1; alternatively -1 and 1) which assigns the input example to the first or the second class. If there are multiple classes (and each example belongs to exactly one of them), it is customary to use one-hot encoding: there are as many outputs as there are classes, with the i-th output equalling 1 if the input example belongs to the i-th class and 0 otherwise. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

21 Binary classification Assume we are building a binary classifier which classifies images into images of dogs and images of cats. Let the "dog" class be encoded as 0 and the "cat" class be encoded as 1. The expected network outputs are shown below the following images: (a) 0 (b) 1 (c) 1 (d) 0 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

22 Multiclass classification Let s look at a classifier that would classify images into three classes: dogs, cats and parrots. We could ask the output to be a one-hot encoded vector: 100 for dog, 010 for cat, 001 for parrot. (a) 100 (b) 010 (c) 010 (d) 001 (e) 100 (f) 001 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

23 Binary classification: example During the hiring procedure for the ACME company: each candidate partakes in two independent interviews and is given a rating of 1 to 5 for each based on the two ratings, the HR manager gives a positive or negative opinion on the candidate candidates that get a positive opinion are directed to the second phase of the hiring procedure In an effort to reduce human effort, part of the procedure is being automated. Specifically, we want an AI system to give a positive/negative opinion instead of the HR manager. To develop such a system a data set of the HR manager s past decisions has been collected. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

24 Binary classification: example Let x 1 be the candidate rating in the first interview and x 2 be the candidate rating in the second interview Company archives contain the following data about ratings and HR manager opinions for past candidates: (x 2, x 1 ) = (2, 5): opinion was positive (x 2, x 1 ) = (5, 2): opinion was positive (x 2, x 1 ) = (1, 5): opinion was negative (x 2, x 1 ) = (5, 1): opinion was negative Based on this data we wish to learn a classifier that would be able to independently determine the appropriate opinion for future candidates. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

25 Binary classification: example Since there are only two classes, as a classifier we will use a single neuron with a step activation function whose output is, depending on th input, either -1 or 1 the class negative opinion will be encoded by -1 the class positive opinion will be encoded by 1 This setup yields the following set of training examples: x 2 x 1 Class t 2 5 positive opinion positive opinion negative opinion negative opinion -1 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

26 Binary classification: example The neuron we use: w 0 x 1 w 1 w 2 net prijenosna funkcija step(net) (-1 ili 1) o x 2 The input x 1 is the rating from the first interview The input x 2 is the rating from the second interview. We compute the accumulated sum net = x 2 w 2 + x 1 w 1 + w 0 The final output is given as o = step(net) Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

27 Binary classification: example Assume we set the weights to random values, and they are w 2 = 1, w 1 = 1.3 i w 0 = How will this neuron classify the examples from the training set? (x 2, x 1, x 0 ) t (w 2, w 1, w 0 ) net o Correct (2, 5, 1) 1 (1, 1.3, 5.85) yes (5, 2, 1) 1 (1, 1.3, 5.85) yes (1, 5, 1) 1 (1, 1.3, 5.85) no (5, 1, 1) 1 (0.96, 1.1, 5.89) no Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

28 Binary classification: example The decision boundary for the TLU neuron is linear: o = step(net) changes value when the sign of net changes therefore, we are interested in when net = 0, which is: x 2 w 2 + x 1 w 1 + w 0 = 0 the is a line in two dimensional space where the axes correspond to x 1 and x 2 : x 2 = w 1 w 2 x 1 w 0 w 2 where w 1 w 2 is the slope of the line This line divides this two dimensional space into two subspaces: one of the contains all examples for which the classifier outputs -1 and the other contains all examples for which the output is 1 In three dimensions this decision boundary would be a plane, in multiple dimensional spaces it is a hyperplane Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

29 Binary classification: example Let s look at our specific example where we compute net = 1 x x The decision boundary is given in the image. Examples for which the classifier assigns the -1 output are in the yellow subspace (e.g. the example (1,1) is there). Examples for which the classifier assigns 1 as the output are in the white subspace (e.g. the example (5,5) is there). The image shows that the classifier is not very good. All training examples are assigned the output 1. x x 1 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

30 Learning from examples Hebb found out by studying biological neurons: to learn is to change the weights of connections between neurons Rosenblatt combines Hebbs idea and the McCulloch-Pitts neuron model and defines the perceptron learning rule Perceptron learning rule 1 Cyclically go through all N training examples, one by one. 2 Classify the current training example 1 If the classification decision is correct, do not change the weights. 1 if this is the N-th consecutive correctly classified example, stop the procedure, 2 otherwise go to the next training example. 2 If the classification decision is incorrect, adapt the weights using the following expression: w i (k + 1) w i (k) + η (t o) x i Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

31 Learning from examples The parameter Parametar η (eta) is called the learning rate It is a small positive number (e.g., between and 0.5) which defines how much the current weight values are going to be modified. If η is too small, the learning procedure will be progress very slowly. If η is too big, the learning procedure could diverge. Let s run the learning procedure with η = 0.02 (refer to the material "Umjetne neuronske mreže" available at chapter 2 for the entire example). The procedure ends with the following weights: w 2 = 0.92, w 1 = 0.94, w 0 = Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

32 Binary classification: example Now the TLU-perceptron computes net = 0.92 x x All training examples are correctly classified The system has learned to generalize: examples like (1,1) will get the label -1 and examples such as (5,5) will get the label 1. This corresponds to decisions a human would make x x 1 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

33 Limitations of the TLU-perceptron the TLU-perceptron has a linear decision boundary Such a neuron can not solve classification problems where the classes are not linearly separable An example of linearly-inseparable classes (the XOR function) is given in the image to the right To solve such more complex problems we will consider systems that are comprised of more neurons: artificial neural networks x x1 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

34 Matrix notation In implementation, especially when dealing with neural networks, it s appropriate to use matrix notation. We will stick to the following convention: a vector will denote a single column matrix as y = f ( x) where f is a scalar function of a single variable will denote a new vector obtained by applying f to every element of x, i.e. y(i) = f (x(i)). If a neuron has n inputs, we denote them as x = (x 1, x 2,..., x n ) In such a case it also has n weights w = (w 1, w 2,..., w n ) and one threshold b (i.e. w 0 ) In this notation: o = f (net) = f ( w T x + b) If we have access to an appropriate library for matrix/vector operations our code can be very concise and efficient. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

35 Contents 1 Introduction to neuro-computing 2 Artificial neuron 3 Artificial neural network Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

36 Architectures To enable modelling of more complex patterns in data for classification and regression, we will use more neurons The architecture of a neural network tells us how many neurons there are and in what way they are connected Layered neural network Non layered neural network x 1 o 1 x 1 o 1 x 2 o 2 x 2 o 2 ulazni sloj 1. skriveni sloj 2. skriveni sloj izlazni sloj ulazni sloj 1. skriveni sloj 2. skriveni sloj izlazni sloj For the network on the right a problem is the recurrent connection which means the network is no longer feedforward. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

37 The need for non-linear activation functions Assume we have built a multi-layered neural network in which all neurons have the identity as the activation function This entire neural network is equally expressive as a single neuron of this type: a linear combination of linear combinations is itself just a linear combination To increase the expressive power of the network and allow the modelling of non-linear patterns, the neurons must have non-linear activation functions. For feedforward neural networks: until recently it was common to use sigmoid activation functions today ReLU functions are preferred because they allow training deeper networks (networks with more layers) Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

38 Feedforward layered neural network: example 1. sloj neurona (1) w 2. sloj neurona 3. sloj neurona 4. sloj neurona sloj težina y 1 (1) w 11 (1) 1 2 (2) y 1 2. sloj težina (2) w 11 (2) w 01 (2) (2) w 13 w sloj težina (3) y 1 (3) (3) w w x y 1 (4) o 1 x 2 2 y 2 (1) 3 2 w 03 (2) w 02 (3) 2 y 2 (4) o 2 w 25 (1) 4 3 y 3 (3) w 32 (3) w 05 (1) w 51 (2) w 52 (2) w 53 (2) 5 y 5 (2) ulazni sloj 1. skriveni sloj 2. skriveni sloj izlazni sloj Slika : Feedforward layered neural network: example Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

39 Feedforward layered neural network: example this network has two hidden layers and one output layer it performs the following mapping: R R R R conventions: we will denote the input of the network with a two component vector: x the output of each subsequent layer will also be a vector of dimensionality equal to the number of neurons in the layer the outputs of hidden layers will be denoted by h, which gives h 1 (dim=5), h 2 (dim=3) and y = h 3 (dim=2) Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

40 Feedforward layered neural network: example The first neuron of the first hidden layer computes: ( [ ] [ ] ) y (2) 1 = f w (1) 11 w (1) x1 + w (1) The second neuron of the first hidden layer computes: ( [ ] [ ] ) y (2) 2 = f w (1) 12 w (1) x1 + w (1) If we use matrix notation the complete output of the layer can be written as: y (2) 1 w (1) y (2) 11 w (1) 21 w (1) 2 y (2) w (1) 12 w (1) [ ] 3 = f w (1) y (2) 13 w (1) x1 w (1) w (1) y (2) 14 w (1) x 2 w (1) w (1) 5 w (1) 15 w (2) w (1) 05 Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56 x 2 x 2

41 Feedforward layered neural network: example A more concise way to write the previous expression: h1 = f (W 1 x + b 1 ) where h 1 is the vector of outputs for the layer, W 1 is the weight matrix (one line represents weights of one neuron), x a vector of inputs, and b 1 a vector of thresholds (biases) with an additional generalization we come to this expression: hi = f (W i h i 1 + b i ) with h 0 = x and y = h 3, which is easy to implement note: for each layer we must know the weight matrix, the thresholds vector and the activation function Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

42 Learning: adapting weights Which outputs the network will give for given inputs is determined by the weights and thresholds of the network During supervised learning, the network has access to a set of training examples consisting of pairs (input, desired_output): {(x 1,1,..., x 1,Ni ) (t 1,1,..., t 1,No ),..., (x N,1,..., x N,Ni ) (t N,1,..., t N,No )} where N i is the dimensionality of the input (number of features) and N o is the dimensionality of the output. To arrive at an algorithm to adapt weights we must first define an error function E. An often used error function is the halved sum of mean squared differences: E = 1 2 N E(s) = 1 2 s=1 N s=1 1 N N o i=1 (t s,i o s,i ) 2. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

43 Learning: adapting weights Once the error function has been defined, if it is differentiable, it s possible to write an optimization procedure based on calculating the gradients (partial derivatives of the error function with respect to each weight and threshold) the partial derivative tells us how the error function will change if we increase/decrease the weight we can use this information to modify the weight in a way that will decrease the error function E Error backpropagation An algorithm for learning neural networks, based on efficient computation of all partial derivatives of the error function and using them to determine the appropriate modification for each weight is called Error Backpropagation We will not cover the derivation, but will only present the final result. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

44 Error backpropagation: the algorithm 1 Initialize weights to random values 2 Repeat until the stopping criterion is satisfied 1 For each training example s : (x s,1,..., x s,ni ) (t s,1,..., t s,no ) do: 1 Set the example (x s,1,..., x s,ni ) as the input of the network. 2 Compute outputs of all neurons from all layers, from the first to the last layer; denote the outputs of the final layer as (o s,1,..., o s,no ). 3 Determine errors of output layer neurons: δ K i = o s,i (1 o s,i ) (t s,i o s,i ). 4 Go back layer by layer toward the first layer. For the i-th neuron of the k-th layer the error is: δ (k) i = y (k) i (1 y (k) i ) w i,d δ (k+1) d 5 Modify all weights. Weight w (k) i,j w (k) i,j w (k) i,j and the thresholds are modified as: d Downstream is modified as: + η y (k) i δ (k+1) j w (k) 0,j w (k) 0,j + η δ (k+1) j. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

45 Error backpropagation: the algorithm In this course we do not cover the derivation of the Backpropagation algorithm A more detailed description of the algorithm is given in the additional materials The expressions in the pseudocode assume that the sigmoid function was used as the activation function for all neurons In case of other transfer functions, parts of expressions which appeared as the derivative of the sigmoid (f (1 f )) must be replaced by the derivatives of the alternative activation function The expressions for weight modifications can be remembered "visually" as shown on the next two slides Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

46 Error backpropagation: the algorithm Error computation for a neuron in the output layer. y i (k) w ij (k) y j (k+1) o j želim: t j izlazni sloj Error: product of the derivation of the neurons activation function (y (k+1) j (1 y (k+1) j )) and the actual error (t j o j ). δ k+1 j = o s,j (1 o s,j ) (t s,j o s,j ). Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

47 Error backpropagation: the algorithm Error computation for a neuron of the hidden layer: δ (k) j. 1 y 1 (k+1) w j1 (k) y i (k-1) w ij (k-1) j y j (k) w j2 (k) 2 y 2 (k+1) w jm (k) m y m (k+1) Error: product of the neurons activation function and a weighted sum of errors of all neurons to which this neuron sends its output: δ (k) j = y (k) j (1 y (k) j ) (w (k) j,1 δ(k+1) w (k) j,m δ(k+1) m ). Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

48 Error backpropagation: the algorithm Weight modification: proportional to the product of the learning rate η, the output of the neuron to the left of the weight and the error of the neurona to the right of the weight: y i (k) w ij (k) y j (k+1) o j želim: t j izlazni sloj wij k = η y (k) i δ (k+1) j wij k wij k + wij k For thresholds the output of the neuron "to the left" is a constant 1, so we have: w k 0j = η δ (k+1) j Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

49 Examples At the address in the section Umjetne neuronske mreže implementations and descriptions of several examples are available. Classification of examples in 2D Functional regression Gesture classification - how to convert a gesture performed with a mouse into a class (digits 0-9) Recommendation: try it out for yourself. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

50 Example: classification and regression We consider using a neural network for classifying 2D examples and for regression. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

51 Example: digit recognition using gestures Task: create a program that uses a neural network to recognize digits (0-9) based on a gesture which the user performs using the mouse The program allows gathering training examples, carrying out the learning procedure, and using the trained network Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

52 Example: classifying bank notes Task: classify four types of paper bank notes regardless of orientation The set of training examples consists of 16 different patterns sampled with 45x22 image elements Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

53 Example: classifying bank notes The examples are preprocessed before putting them at the input of the ANN Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

54 Example: classifying bank notes Since the examples are digitalized images, they will be different in the intensity of corresponding image elements (a consequence of differences in the general degree of wear and weather the paper is crumpled or otherwise damaged) In the example, a generator of artificial examples is used, which generates examples with varying degrees of wear, in order to validate whether the network works and tune its parameters. Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

55 Example: classifying bank notes Network parameters: acyclic fully connected multi-layered neural network Learning: error backpropagation, learning_rate=0.02, moment=0.02, validating the model on the validation set every 2500 epochs Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

56 What s next? Artificial neural networks nowadays have numerous applications (processing of images, sound, text) Some types of networks we omitted today are also often used: convolutional neural networks, recurrent neural networks More information is available in graduate level courses Machine learning Fuzzy, evolutionary, and neuro-computing Deep learning Dalbelo Bašić, Čupić, Šnajder (FER) Artificial Neural Networks lipanj / 56

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

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 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

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

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

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

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. 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

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

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

Introduction to Neural Networks

Introduction to Neural Networks Introduction to Neural Networks What are (Artificial) Neural Networks? Models of the brain and nervous system Highly parallel Process information much more like the brain than a serial computer Learning

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

Course 395: Machine Learning - Lectures

Course 395: Machine Learning - Lectures Course 395: Machine Learning - Lectures Lecture 1-2: Concept Learning (M. Pantic) Lecture 3-4: Decision Trees & CBC Intro (M. Pantic & S. Petridis) Lecture 5-6: Evaluating Hypotheses (S. Petridis) Lecture

More information

Neural Networks (Part 1) Goals for the lecture

Neural Networks (Part 1) Goals for the lecture Neural Networks (Part ) Mark Craven and David Page Computer Sciences 760 Spring 208 www.biostat.wisc.edu/~craven/cs760/ Some of the slides in these lectures have been adapted/borrowed from materials developed

More information

Neural Networks: Introduction

Neural Networks: Introduction Neural Networks: Introduction Machine Learning Fall 2017 Based on slides and material from Geoffrey Hinton, Richard Socher, Dan Roth, Yoav Goldberg, Shai Shalev-Shwartz and Shai Ben-David, and others 1

More information

Simple Neural Nets For Pattern Classification

Simple Neural Nets For Pattern Classification CHAPTER 2 Simple Neural Nets For Pattern Classification Neural Networks General Discussion One of the simplest tasks that neural nets can be trained to perform is pattern classification. In pattern classification

More information

DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY

DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY DEEP LEARNING AND NEURAL NETWORKS: BACKGROUND AND HISTORY 1 On-line Resources http://neuralnetworksanddeeplearning.com/index.html Online book by Michael Nielsen http://matlabtricks.com/post-5/3x3-convolution-kernelswith-online-demo

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. 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

CSC321 Lecture 5: Multilayer Perceptrons

CSC321 Lecture 5: Multilayer Perceptrons CSC321 Lecture 5: Multilayer Perceptrons Roger Grosse Roger Grosse CSC321 Lecture 5: Multilayer Perceptrons 1 / 21 Overview Recall the simple neuron-like unit: y output output bias i'th weight w 1 w2 w3

More information

Apprentissage, réseaux de neurones et modèles graphiques (RCP209) Neural Networks and Deep Learning

Apprentissage, réseaux de neurones et modèles graphiques (RCP209) Neural Networks and Deep Learning Apprentissage, réseaux de neurones et modèles graphiques (RCP209) Neural Networks and Deep Learning Nicolas Thome Prenom.Nom@cnam.fr http://cedric.cnam.fr/vertigo/cours/ml2/ Département Informatique Conservatoire

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

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

Artificial Neural Networks

Artificial Neural Networks Artificial Neural Networks 鮑興國 Ph.D. National Taiwan University of Science and Technology Outline Perceptrons Gradient descent Multi-layer networks Backpropagation Hidden layer representations Examples

More information

Single layer NN. Neuron Model

Single layer NN. Neuron Model Single layer NN We consider the simple architecture consisting of just one neuron. Generalization to a single layer with more neurons as illustrated below is easy because: M M The output units are independent

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

Artificial Neural Networks Examination, March 2004

Artificial Neural Networks Examination, March 2004 Artificial Neural Networks Examination, March 2004 Instructions There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the mark obtained in the laborations (maximum

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

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

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

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

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

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

Artificial Neural Networks. Edward Gatt

Artificial Neural Networks. Edward Gatt Artificial Neural Networks Edward Gatt What are Neural Networks? Models of the brain and nervous system Highly parallel Process information much more like the brain than a serial computer Learning Very

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

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

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

Nonlinear Classification

Nonlinear Classification Nonlinear Classification INFO-4604, Applied Machine Learning University of Colorado Boulder October 5-10, 2017 Prof. Michael Paul Linear Classification Most classifiers we ve seen use linear functions

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

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

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

Chapter 2 Single Layer Feedforward Networks

Chapter 2 Single Layer Feedforward Networks Chapter 2 Single Layer Feedforward Networks By Rosenblatt (1962) Perceptrons For modeling visual perception (retina) A feedforward network of three layers of units: Sensory, Association, and Response Learning

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

Artificial Neural Networks Examination, June 2004

Artificial Neural Networks Examination, June 2004 Artificial Neural Networks Examination, June 2004 Instructions There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the mark obtained in the laborations (maximum

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

The Perceptron. Volker Tresp Summer 2016

The Perceptron. Volker Tresp Summer 2016 The Perceptron Volker Tresp Summer 2016 1 Elements in Learning Tasks Collection, cleaning and preprocessing of training data Definition of a class of learning models. Often defined by the free model parameters

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

Reification of Boolean Logic

Reification of Boolean Logic 526 U1180 neural networks 1 Chapter 1 Reification of Boolean Logic The modern era of neural networks began with the pioneer work of McCulloch and Pitts (1943). McCulloch was a psychiatrist and neuroanatomist;

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

Artificial Neural Network

Artificial Neural Network Artificial Neural Network Eung Je Woo Department of Biomedical Engineering Impedance Imaging Research Center (IIRC) Kyung Hee University Korea ejwoo@khu.ac.kr Neuron and Neuron Model McCulloch and Pitts

More information

Neural Networks with Applications to Vision and Language. Feedforward Networks. Marco Kuhlmann

Neural Networks with Applications to Vision and Language. Feedforward Networks. Marco Kuhlmann Neural Networks with Applications to Vision and Language Feedforward Networks Marco Kuhlmann Feedforward networks Linear separability x 2 x 2 0 1 0 1 0 0 x 1 1 0 x 1 linearly separable not linearly separable

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

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

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

The Perceptron. Volker Tresp Summer 2014

The Perceptron. Volker Tresp Summer 2014 The Perceptron Volker Tresp Summer 2014 1 Introduction One of the first serious learning machines Most important elements in learning tasks Collection and preprocessing of training data Definition of a

More information

Neural Networks and Deep Learning

Neural Networks and Deep Learning Neural Networks and Deep Learning Professor Ameet Talwalkar November 12, 2015 Professor Ameet Talwalkar Neural Networks and Deep Learning November 12, 2015 1 / 16 Outline 1 Review of last lecture AdaBoost

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

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

PV021: Neural networks. Tomáš Brázdil

PV021: Neural networks. Tomáš Brázdil 1 PV021: Neural networks Tomáš Brázdil 2 Course organization Course materials: Main: The lecture Neural Networks and Deep Learning by Michael Nielsen http://neuralnetworksanddeeplearning.com/ (Extremely

More information

Deep Learning: a gentle introduction

Deep Learning: a gentle introduction Deep Learning: a gentle introduction Jamal Atif jamal.atif@dauphine.fr PSL, Université Paris-Dauphine, LAMSADE February 8, 206 Jamal Atif (Université Paris-Dauphine) Deep Learning February 8, 206 / Why

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

Artificial Neural Networks D B M G. Data Base and Data Mining Group of Politecnico di Torino. Elena Baralis. Politecnico di Torino

Artificial Neural Networks D B M G. Data Base and Data Mining Group of Politecnico di Torino. Elena Baralis. Politecnico di Torino Artificial Neural Networks Data Base and Data Mining Group of Politecnico di Torino Elena Baralis Politecnico di Torino Artificial Neural Networks Inspired to the structure of the human brain Neurons as

More information

Need for Deep Networks Perceptron. Can only model linear functions. Kernel Machines. Non-linearity provided by kernels

Need for Deep Networks Perceptron. Can only model linear functions. Kernel Machines. Non-linearity provided by kernels Need for Deep Networks Perceptron Can only model linear functions Kernel Machines Non-linearity provided by kernels Need to design appropriate kernels (possibly selecting from a set, i.e. kernel learning)

More information

Machine Learning

Machine Learning Machine Learning 10-315 Maria Florina Balcan Machine Learning Department Carnegie Mellon University 03/29/2019 Today: Artificial neural networks Backpropagation Reading: Mitchell: Chapter 4 Bishop: Chapter

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

Artificial Neural Networks Artificial Neural Networks CPSC 533 Winter 2 Christian Jacob Neural Networks in the Context of AI Systems Neural Networks as Mediators between Symbolic AI and Statistical Methods 2 5.-NeuralNets-2.nb Neural

More information

Need for Deep Networks Perceptron. Can only model linear functions. Kernel Machines. Non-linearity provided by kernels

Need for Deep Networks Perceptron. Can only model linear functions. Kernel Machines. Non-linearity provided by kernels Need for Deep Networks Perceptron Can only model linear functions Kernel Machines Non-linearity provided by kernels Need to design appropriate kernels (possibly selecting from a set, i.e. kernel learning)

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

Fundamentals of Neural Networks

Fundamentals of Neural Networks Fundamentals of Neural Networks : Soft Computing Course Lecture 7 14, notes, slides www.myreaders.info/, RC Chakraborty, e-mail rcchak@gmail.com, Aug. 10, 2010 http://www.myreaders.info/html/soft_computing.html

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

Using a Hopfield Network: A Nuts and Bolts Approach

Using a Hopfield Network: A Nuts and Bolts Approach Using a Hopfield Network: A Nuts and Bolts Approach November 4, 2013 Gershon Wolfe, Ph.D. Hopfield Model as Applied to Classification Hopfield network Training the network Updating nodes Sequencing of

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

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

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

In the Name of God. Lecture 9: ANN Architectures

In the Name of God. Lecture 9: ANN Architectures In the Name of God Lecture 9: ANN Architectures Biological Neuron Organization of Levels in Brains Central Nervous sys Interregional circuits Local circuits Neurons Dendrite tree map into cerebral cortex,

More information

Artificial Neural Networks. Q550: Models in Cognitive Science Lecture 5

Artificial Neural Networks. Q550: Models in Cognitive Science Lecture 5 Artificial Neural Networks Q550: Models in Cognitive Science Lecture 5 "Intelligence is 10 million rules." --Doug Lenat The human brain has about 100 billion neurons. With an estimated average of one thousand

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

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

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

The perceptron learning algorithm is one of the first procedures proposed for learning in neural network models and is mostly credited to Rosenblatt.

The perceptron learning algorithm is one of the first procedures proposed for learning in neural network models and is mostly credited to Rosenblatt. 1 The perceptron learning algorithm is one of the first procedures proposed for learning in neural network models and is mostly credited to Rosenblatt. The algorithm applies only to single layer models

More information

Artificial Neural Networks. Part 2

Artificial Neural Networks. Part 2 Artificial Neural Netorks Part Artificial Neuron Model Folloing simplified model of real neurons is also knon as a Threshold Logic Unit x McCullouch-Pitts neuron (943) x x n n Body of neuron f out Biological

More information

Artificial Intelligence (AI) Common AI Methods. Training. Signals to Perceptrons. Artificial Neural Networks (ANN) Artificial Intelligence

Artificial Intelligence (AI) Common AI Methods. Training. Signals to Perceptrons. Artificial Neural Networks (ANN) Artificial Intelligence Artificial Intelligence (AI) Artificial Intelligence AI is an attempt to reproduce intelligent reasoning using machines * * H. M. Cartwright, Applications of Artificial Intelligence in Chemistry, 1993,

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

(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

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

CSCI567 Machine Learning (Fall 2018)

CSCI567 Machine Learning (Fall 2018) CSCI567 Machine Learning (Fall 2018) Prof. Haipeng Luo U of Southern California Sep 12, 2018 September 12, 2018 1 / 49 Administration GitHub repos are setup (ask TA Chi Zhang for any issues) HW 1 is due

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

Neural Nets in PR. Pattern Recognition XII. Michal Haindl. Outline. Neural Nets in PR 2

Neural Nets in PR. Pattern Recognition XII. Michal Haindl. Outline. Neural Nets in PR 2 Neural Nets in PR NM P F Outline Motivation: Pattern Recognition XII human brain study complex cognitive tasks Michal Haindl Faculty of Information Technology, KTI Czech Technical University in Prague

More information

The Perceptron. Volker Tresp Summer 2018

The Perceptron. Volker Tresp Summer 2018 The Perceptron Volker Tresp Summer 2018 1 Elements in Learning Tasks Collection, cleaning and preprocessing of training data Definition of a class of learning models. Often defined by the free model parameters

More information

Classification with Perceptrons. Reading:

Classification with Perceptrons. Reading: Classification with Perceptrons Reading: Chapters 1-3 of Michael Nielsen's online book on neural networks covers the basics of perceptrons and multilayer neural networks We will cover material in Chapters

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

Master Recherche IAC TC2: Apprentissage Statistique & Optimisation

Master Recherche IAC TC2: Apprentissage Statistique & Optimisation Master Recherche IAC TC2: Apprentissage Statistique & Optimisation Alexandre Allauzen Anne Auger Michèle Sebag LIMSI LRI Oct. 4th, 2012 This course Bio-inspired algorithms Classical Neural Nets History

More information

Multilayer Neural Networks

Multilayer Neural Networks Multilayer Neural Networks Multilayer Neural Networks Discriminant function flexibility NON-Linear But with sets of linear parameters at each layer Provably general function approximators for sufficient

More information

SGD and Deep Learning

SGD and Deep Learning SGD and Deep Learning Subgradients Lets make the gradient cheating more formal. Recall that the gradient is the slope of the tangent. f(w 1 )+rf(w 1 ) (w w 1 ) Non differentiable case? w 1 Subgradients

More information

Lecture 5 Neural models for NLP

Lecture 5 Neural models for NLP CS546: Machine Learning in NLP (Spring 2018) http://courses.engr.illinois.edu/cs546/ Lecture 5 Neural models for NLP Julia Hockenmaier juliahmr@illinois.edu 3324 Siebel Center Office hours: Tue/Thu 2pm-3pm

More information

COMP304 Introduction to Neural Networks based on slides by:

COMP304 Introduction to Neural Networks based on slides by: COMP34 Introduction to Neural Networks based on slides by: Christian Borgelt http://www.borgelt.net/ Christian Borgelt Introduction to Neural Networks Motivation: Why (Artificial) Neural Networks? (Neuro-)Biology

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

Administration. Registration Hw3 is out. Lecture Captioning (Extra-Credit) Scribing lectures. Questions. Due on Thursday 10/6

Administration. Registration Hw3 is out. Lecture Captioning (Extra-Credit) Scribing lectures. Questions. Due on Thursday 10/6 Administration Registration Hw3 is out Due on Thursday 10/6 Questions Lecture Captioning (Extra-Credit) Look at Piazza for details Scribing lectures With pay; come talk to me/send email. 1 Projects Projects

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. 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