Dropout is a regularisation technique for neural network training in which a randomly selected fraction of neuron activations is set to zero during each forward pass, preventing neurons from co-adapting and forcing the network to learn redundant representations. By randomly deactivating 20–50% of units per training step, dropout acts as an ensemble method — each mini-batch trains a slightly different network architecture — significantly reducing overfitting on limited training datasets. At inference time, all neurons are active but their outputs are scaled by the retention probability.
Semantic Classification
Content
-
Dropout is a regularisation technique that randomly deactivates (drops) a fraction of neurons during each training iteration, preventing co-adaptation and reducing overfitting. Dropout rate (typically 0.2–0.5) controls the fraction of neurons dropped.
Standards Alignment
ISO/IEC Standards
-
IEC 22989:2022: Training and regularisation
NIST AI RMF
-
Function: MAP (Training techniques)
Related Terms
-
Regularisation (AI-0056): Parent category
-
Overfitting (AI-0054): Prevented by dropout
-
Training (AI-0041): Applies dropout
-
Neural Network: Uses dropout layers
References
- Srivastava et al. - “Dropout: A Simple Way to Prevent Neural Networks from Overfitting” - JMLR, 2014
- IEC 22989:2022 - Training techniques
- NIST AI RMF - MAP function mapping