Overview

Hyperparameters are the 'knobs' you turn before training starts, such as the learning rate, batch size, number of layers, or dropout rate. They significantly impact the model's performance.

Methods

  • Grid Search: Trying every possible combination from a predefined list.
  • Random Search: Trying random combinations (often more efficient than grid search).
  • Bayesian Optimization: Using a probabilistic model to find the best settings more quickly.

Goal

To maximize the model's accuracy and generalization while minimizing training time and resource use.

Related Terms