Artificial neural network

Back to Data-Science

Artificial neural networks (ANNs) are simple mathematical models defining f: X -> Y or a distribution over X or both X and Y

Uses a connectionism model, based on modelling mental phenomena using interconnected units. Neural Networks are a computational approach based on a collection of neural units connected by axons.


A single neuron takes input from other nodes and computes an output

Bias lets you shift the activation function left or right, which may be critical for success

adjusting weighting vs. adjusting bias

Feedforward Neural Networks are the simplest type of NN

Convolutional Neural Networks


A category of Neural Networks, effective in image recognition and classification

Feed-forward neural network: connections between units are acyclic, and is unidirectional

ImageNet classification with deep convolutional neural networks

CNN's capacity can be controlled by varying their depth and breadth (layers), and make strong assumptions with less needed connections

Maxout networks are designed to work with dropout networks