Timezone: »

 
Poster
Sharpness-Aware Training for Free
JIAWEI DU · Daquan Zhou · Jiashi Feng · Vincent Tan · Joey Tianyi Zhou

Tue Nov 29 09:00 AM -- 11:00 AM (PST) @ Hall J #221
in Poster Session 1 »

Modern deep neural networks (DNNs) have achieved state-of-the-art performances but are typically over-parameterized. The over-parameterization may result in undesirably large generalization error in the absence of other customized training strategies. Recently, a line of research under the name of Sharpness-Aware Minimization (SAM) has shown that minimizing a sharpness measure, which reflects the geometry of the loss landscape, can significantly reduce the generalization error. However, SAM-like methods incur a two-fold computational overhead of the given base optimizer (e.g. SGD) for approximating the sharpness measure. In this paper, we propose Sharpness-Aware Training for Free, or SAF, which mitigates the sharp landscape at almost zero additional computational cost over the base optimizer. Intuitively, SAF achieves this by avoiding sudden drops in the loss in the sharp local minima throughout the trajectory of the updates of the weights. Specifically, we suggest a novel trajectory loss, based on the KL-divergence between the outputs of DNNs with the current weights and past weights, as a replacement of the SAM's sharpness measure. This loss captures the rate of change of the training loss along the model's update trajectory. By minimizing it, SAF ensures the convergence to a flat minimum with improved generalization capabilities. Extensive empirical results show that SAF minimizes the sharpness in the same way that SAM does, yielding better results on the ImageNet dataset with essentially the same computational cost as the base optimizer.

Keywords: generalization · Efficient learning · training algorithm
[ Paper]  [ Poster]  [ OpenReview] 

Author Information

JIAWEI DU (CENTRE FOR FRONTIER AI RESEARCH (CFAR) A*STAR, national university of singapore)
Daquan Zhou (National University of Singapore)
Jiashi Feng (UC Berkeley)
Vincent Tan (National University of Singapore)
Joey Tianyi Zhou (IHPC, A*STAR)

More from the Same Authors