TR2020-132
Autoclip: Adaptive Gradient Clipping For Source Separation Networks
-
- "Autoclip: Adaptive Gradient Clipping For Source Separation Networks", IEEE International Workshop on Machine Learning for Signal Processing (MLSP), DOI: 10.1109/MLSP49062.2020.9231926, September 2020.BibTeX TR2020-132 PDF
- @inproceedings{Seetharaman2020sep,
- author = {Seetharaman, Prem and Wichern, Gordon and Pardo, Bryan and Le Roux, Jonathan},
- title = {Autoclip: Adaptive Gradient Clipping For Source Separation Networks},
- booktitle = {IEEE International Workshop on Machine Learning for Signal Processing (MLSP)},
- year = 2020,
- month = sep,
- publisher = {IEEE},
- doi = {10.1109/MLSP49062.2020.9231926},
- url = {https://www.merl.com/publications/TR2020-132}
- }
,
- "Autoclip: Adaptive Gradient Clipping For Source Separation Networks", IEEE International Workshop on Machine Learning for Signal Processing (MLSP), DOI: 10.1109/MLSP49062.2020.9231926, September 2020.
-
MERL Contacts:
-
Research Areas:
Abstract:
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