convolutional

Definition

Describes a layer that slides a small learned filter across the input, applying the same weights at every position. This weight sharing bakes in locality and translation equivariance, which is why convolutional networks dominated image recognition for years and remain common in lightweight, audio, and time-series models. Transformers have displaced them in many settings, but the inductive bias still pays off when data or compute is limited.

Explainers using this term

• Graph Neural Networks from Scratch — Learning from ConnectionsSemi-Supervised Classification with Graph Convolutional Networks
• Symmetry and Equivariance — How Group Theory Shapes Network DesignGroup Equivariant Convolutional Networks
• The CNN Family Tree — From AlexNet to ResNet and EfficientNetVery Deep Convolutional Networks for Large-Scale Image Recognition