Neural Networks
Last updated
Was this helpful?
Last updated
Was this helpful?
A neural network is a type of machine learning model loosely inspired by biological neural networks in the human brain. They are designed to recognise patterns and features in data.
At a very basic level, they comprise the following components:
Inputs
Data such as images, text or sound samples get fed into the model
Each input gets assigned to one of the initial nodes
Nodes
Also called neurons or units
Organised in layers - usually an input layer, one or more hidden layers, and an output layer
Each node assigns a weight to its input, performs a calculation on it and passes the value to connected nodes
Think of a node receiving multiple inputs, assessing their importance, and communicating its assessment to other neurons deeper in the network
Connections
Direct connections between the output of nodes in one layer to the input of nodes in the next layer
Enable information and weighted values to be passed along for further processing steps
Outputs
The end result showing what the neural network has predicted based on all the calculations stepped through each layer of connections
Output could be a classification, a prediction, recommendation or other inferred pattern in the data
The strengths of connections between nodes and the node weight values are optimised during model training to improve accuracy on sample data. The trained model can then be applied to new unseen data.
Various neural network types cater to specific tasks, such as vision, numerical processing, and language understanding. In this context, we will provide a brief overview of a large language model (LLM).
