Regularisation in Machine Learning
Regularisation is any technique that discourages a model from becoming overly complex, so it generalises better to new data.
What is Regularisation in Machine Learning?
Regularisation trades a little training accuracy for better real-world performance. Common forms include adding a penalty on large weights (L1 or L2), dropout, data augmentation and early stopping.
How it works
A weight penalty adds a term to the loss function that grows with the size of the weights, nudging the model towards simpler solutions.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions