Unsupervised learning

Unsupervised learning is a framework in <a href="/facts/Machine_learning/e0w0XJTu">machine learning</a> where, in contrast to <a href="/facts/Supervised_learning/JlNFNPFt">supervised learning</a>, algorithms learn patterns exclusively from unlabeled data. Other frameworks in the spectrum of supervisions include <a href="/facts/Weak_supervision/3EAoE32F">weak- or semi-supervision</a>, where a small portion of the data is tagged, and <a href="/facts/Self-supervised_learning/51WjBKrn">self-supervision</a>. Some researchers consider self-supervised learning a form of unsupervised learning.
Conceptually, unsupervised learning divides into the aspects of data, training, algorithm, and downstream applications. Typically, the dataset is harvested cheaply "in the wild", such as massive <a href="/facts/Text_corpus/bYeEQQdh">text corpus</a> obtained by <a href="/facts/Web_crawler/ra7D1xyb">web crawling</a>, with only minor filtering (such as <a href="/facts/Common_Crawl/z0k2mxv3">Common Crawl</a>). This compares favorably to supervised learning, where the dataset (such as the <a href="/facts/ImageNet/DPaK4GLC">ImageNet1000</a>) is typically constructed manually, which is much more expensive.
There were algorithms designed specifically for unsupervised learning, such as <a href="/facts/Cluster_analysis/HUeDnWPt">clustering algorithms</a> like <a href="/facts/K-means_clustering/vxIcqHwN">k-means</a>, <a href="/facts/Dimensionality_reduction/XoRpcrB0">dimensionality reduction</a> techniques like <a href="/facts/Principal_component_analysis/pCbVTqdR">principal component analysis (PCA)</a>, <a href="/facts/Boltzmann_machine/2wyLI0pI">Boltzmann machine learning</a>, and <a href="/facts/Autoencoder/jnTqhtm9">autoencoders</a>. After the rise of deep learning, most large-scale unsupervised learning have been done by training general-purpose neural network architectures by <a href="/facts/Gradient_descent/pFFrek0F">gradient descent</a>, adapted to performing unsupervised learning by designing an appropriate training procedure.
Sometimes a trained model can be used as-is, but more often they are modified for downstream applications. For example, the generative pretraining method trains a model to generate a textual dataset, before finetuning it for other applications, such as text classification. As another example, autoencoders are trained to <a href="/facts/Feature_learning/6GYV0mux">good features</a>, which can then be used as a module for other models, such as in a <a href="/facts/Latent_diffusion_model/RgyDnrNN">latent diffusion model</a>.

Unsupervised learning open-in-new

Unsupervised learning