Neural network — Glossary

A neural network is a parameterized computational graph inspired loosely by biological neurons but implemented as linear algebra and nonlinear functions. Each layer transforms representations; learning adjusts connection weights to reduce a loss on training data.

Networks range from small multilayer perceptrons to large transformers with billions of parameters. Their effectiveness depends on architecture, data quality, regularization, and optimization. Interpretability varies: some structure can be probed, but many behaviors remain hard to explain fully.