Researchers began proposing the Ising magnetic model for cognitive tasks in 1974. This early work laid the groundwork for understanding how machines might learn without explicit labels.
John Hopfield described an Ising variant known as the Hopfield net in 1982. Kunihiko Fukushima introduced the neocognitron in 1980, later recognized as a convolutional neural network.
Supervised methods rely on tagged datasets like ImageNet1000 while unsupervised approaches harvest information cheaply from sources such as Common Crawl. The dataset is often massive and requires only minor filtering before analysis begins.
Restricted Boltzmann Machines separate into two layers: hidden versus visible. They maintain symmetric two-way weights but prohibit lateral connections within any single layer.
Cluster analysis uses K-means algorithms to identify commonalities in data and react based on their presence or absence. Hierarchical clustering and DBSCAN offer alternative methods for segmenting information.