We observed a clear trend: increasing the weight decay (the hyperparameter controlling L2 regularisation) improved the performance of the classical CNN components.
← all excerpts
Representation learning with parameterised quantum circuits for advancing speech emotion recognition.
1
—
—