Artificial Neural Network Models Overview

Study Notes

Artificial neural network (ANN) models are inspired by the structure of the human brain.
ANNs consist of interconnected nodes or neurons arranged in layers to process information.
Different types of ANN architectures offer unique functionalities and applications.

In feedforward ANNs, data flows in one direction - from input to output nodes.
The information passes through layers of nodes, and there's no backward flow or looping.

Perceptron: A simple binary classifier, separating data into two categories.
Multilayer Perceptron (MLP): Adds complexity and depth, with multiple hidden layers between input and output.
MLPs can handle more complex data and classifications.

RBF networks typically have three layers: input, radial basis function (RBF) layer, and output layer.
RBF nodes have different parameters and are used in diverse applications like classification, regression, and control systems.
RBF functions calculate the distance between a center point (or prototype) and a given data point.
In classification, RBFs determine the distance between an input and learned classifications.
The input closer to a specific prototype is classified accordingly.

RNNs are well-suited for processing sequential data, like speech or text.
Information flows forward through layers and also loops back to previous steps, enabling the network to "remember" past information.
This allows for enhanced prediction capabilities and handling of time-dependent data.
RNNs are instrumental in sequence-to-sequence models used for natural language processing (NLP) tasks.

Modular ANNs combine multiple smaller networks or modules.
Each module handles a specific task, and their collaborative effort leads to a complex task completion.
This approach promotes modularity, flexibility, and efficiency in complex systems.
Decomposition of tasks into smaller components allows parallel processing, leading to faster and more efficient computations.