Introduction to Neural Networks Neural Network Architecture and Learning
Neural Networks Notations
Lecture Overview The Neuron Model Neural Networks Architectures Learning and Learning Rules in Neural Networks.
Introduction Artificial neuron models and neural network architectures are generally quite simple. Why are artificial neural networks so powerful in solving computationally complex tasks, such as image/speech understanding, optimisation, and process control? This is because their ability to learn and to generalise.
Neuron Model
Activation Function
Neural Networks Architectures
Neural Networks Architecture
How to Design a Neural Network for a Given Task A neural network model consists of : (1)a neuron model (2)interconnection architecture (3)A learning (optimisation) algorithm. The design process includes: (1)Choosing a neuron model (2)Selecting a neural network architecture (3)Collecting training data and developing a learning algorithm (4)Training the network and testing its performance The neural network performance is evaluated by: –Convergence/stability –Accuracy –Storage capacity –Generalisation.
Learning in Neural Networks Learning is a process by which the free parameters (including hyper-parameters controlling the network structure) of a neural network are adapted through a process of stimulation by the environment where the network is embedded. The type of learning is determined by the manner in which the parameter changes take place.
Learning Tasks and Performance Measures
Hebbian Learning Rule-The oldest rule
Learning in Neural Networks
Hebbian Learning Rule
Competitive Learning Rule
Error Correction Learning Rule
Performance Improvement through Error Correction Learning
Error Gradient Descent Learning Rule