Neural network

A computational model made of connected units with adjustable parameters. It learns patterns by changing those parameters in response to data or an objective.

Artificial neuron: A computational unit that combines inputs, applies parameters and an activation function, and produces an output. Each unit transforms signals passed through a network.
Training data: Examples used to fit a model’s adjustable parameters. Networks adapt their parameters using patterns present in these examples.
Convolutional neural network: A neural network architecture that uses convolutional operations to process grid-like data such as images. Shared filters help networks detect spatial patterns in images.
Decision tree: A predictive model that partitions data through a sequence of feature-based decisions. Unlike many neural networks, its prediction path can be read as explicit rules.
Activation function: A function applied to a unit’s weighted input to determine its output. Nonlinear activations let networks represent more than linear relationships.
Model parameter: A value learned or adjusted during model fitting, such as a neural network weight or bias. Weights and biases determine how strongly units influence later computations.
Transformer (deep learning architecture): A neural network architecture that uses attention mechanisms to process relationships among sequence elements. Transformers underpin many current language and multimodal systems.
Support vector machine: A supervised learning method that separates classes by optimizing a decision boundary and margin. It offers a distinct approach to classification, often effective on smaller datasets.
Backpropagation: An algorithm for computing how a loss changes with a model’s parameters by propagating derivatives backward through a computation graph. It supplies gradients used to adjust network parameters during training.
Feature (machine learning): A measurable input variable used by a machine-learning model to make predictions. Features provide the input signals a network learns to interpret.

Linked from 25 pages