Can we tune the filter size and number of filters in a CNN automatically using Bayesian optimization?

6 views (last 30 days)
what are all the hyper-parameters that can be tuned using Bayesian optimization in experiment manager?

Accepted Answer

Hrishikesh Borate
Hrishikesh Borate on 15 Jul 2021
The hyperparameters to be tuned can be added in the Experiment Manager. In the code file, which contains the network definition, these hyperparameters can be accessed via the params variable, which is a structure with fields from the Experiment Manager hyperparameter table. These hyperparameters should be declared in the network definition code before starting the process of hyperparameter tuning.
For example, to tune the filter size hyperparameter in this example, first add this hyperparameter in the Experiment Manager. Define the Range, Type, Transform for this hyperparameter.
Now, use this hyperparameter in the network definition (defined in the BayesOptExperiment_setup1.mlx file).
After all the hyperparameters to be tuned are added in both, the Experiment Manager and in the code file, the hyperparameter tuning process can commence.
For more information, refer to the Tune Experiment Hyperparameters by Using Bayesian Optimization documentation.

More Answers (0)


Find more on Optimization Toolbox in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by