Perceptron

Perceptron Pattern Recognition 1. Letter Recognition 2. Image Recognition 3. Disease Diagnosis 4. Stock Price Prediction 5. Business Failure Prediction Artificial Neural Network Structure Artificial Neural Network consists of 4 structures, namely: 1. Single Layer Fully Connected 2. Two Layer Feedforward 3. Multilayer Feedforward 4. Organized Feature Map Perceptron Perceptron is a simple artificial neural network architecture consisting of one layer of input units (whose number of neurons corresponds to the number of components of the data to be recognized) and one output unit. Perceptron is a synonym for Single-Layer, Feed- Forward Network. The term was first studied in the 50s. The First Neural Networks 1. AND Function 2. OR Function 3. AND – NOT Function 4. XOR Function X1 XOR X2 = (X1 AND NOT X2) OR (X2 AND NOT X1) What Can Perceptron Represent? The function that separates the region into the above is known as Linearly Separable. Only linearly separable functions can be represented by a perceptron. Perceptron Learning Perceptron learning assumes that the learning procedure is proven to lead the weights to converge. The fundamental difference with Hebb net is that in perceptron there is an error in the input pattern that is trained, then the weight will change according to the formula: w i (new) = w i (old) +  tx i If there is no error then the weight value will not change, or in other words, the weight only changes following certain conditions. Algorithms 1. Step 0: Initialize all weights and biases. For simplicity, set weights and bias = 0. Set learning rate (0    1) For simplicity, set =1 2. Step 1: as long as the stop condition is not met, perform steps 2 – 6 3. Step 2: for each pair of input and target output vectors, s: t, perform steps 3 - 5. 4. Step 3: Input activation set: x i = s i , 5. Step 4: Calculate the output: y_in = b +i x i w i y = 1 if y_in >  y = 0 if -   y_in   y = -1 if y_in <  6. Step 5: update the weight and bias values if there is an error 7. Step 6: Stop condition test If no weight has changed in step 2, then stop, otherwise continue.