AutoClip: Adaptive Gradient Clipping for Source Separation Networks
This work addresses the hyperparameter tuning challenge for gradient clipping in source separation, offering a simple, domain-agnostic solution that is incremental but practical.
The authors tackled the problem of manually selecting gradient clipping thresholds in optimization by introducing AutoClip, an adaptive method that automatically adjusts the threshold based on historical gradient norms, resulting in improved generalization performance for audio source separation networks.
Clipping the gradient is a known approach to improving gradient descent, but requires hand selection of a clipping threshold hyperparameter. We present AutoClip, a simple method for automatically and adaptively choosing a gradient clipping threshold, based on the history of gradient norms observed during training. Experimental results show that applying AutoClip results in improved generalization performance for audio source separation networks. Observation of the training dynamics of a separation network trained with and without AutoClip show that AutoClip guides optimization into smoother parts of the loss landscape. AutoClip is very simple to implement and can be integrated readily into a variety of applications across multiple domains.