AI Dictionary › AI Fundamentals

Weight Decay

Weight decay is a regularization technique that penalizes overly large weights during training, adding a term proportional to their magnitude to the loss function. The goal is to keep the model's weights contained, favoring simpler solutions and reducing the risk of overfitting. It is one of the most common hyperparameters configured when training neural networks.

Definition

How it works

During each update, in addition to pushing weights toward values that reduce error on the data, the optimizer slightly shrinks their magnitude, like a constant force pulling them toward zero. The strength of this penalty is controlled by a coefficient that balances fitting the data against keeping the model simple.

Applications

It is widely used when training deep neural networks, often combined with other regularization methods such as dropout and early stopping. It is built into many modern optimizers, including variants specifically designed to apply it more effectively alongside adaptive learning-rate updates.

History & etymology

It is a regularization technique with roots in classical statistical methods for penalizing parameters, long adopted in neural network training.

Related terms

More in AI Fundamentals

Put it into practice

From our network

INDACO TMS: Transport Management for European Logistics

Shipment tracking, multi-carrier EDI and automated invoicing in one cloud platform. Invoices generated in under 10 seconds.

Visit indacotms.com →

From the Agora Intelligence blog

More on agora-intelligence.com →

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.