papers AI Learner
The Github is limit! Click to go to the new site.

Essentially No Barriers in Neural Network Energy Landscape

2019-02-22
Felix Draxler, Kambis Veschgini, Manfred Salmhofer, Fred A. Hamprecht

Abstract

Training neural networks involves finding minima of a high-dimensional non-convex loss function. Knowledge of the structure of this energy landscape is sparse. Relaxing from linear interpolations, we construct continuous paths between minima of recent neural network architectures on CIFAR10 and CIFAR100. Surprisingly, the paths are essentially flat in both the training and test landscapes. This implies that neural networks have enough capacity for structural changes, or that these changes are small between minima. Also, each minimum has at least one vanishing Hessian eigenvalue in addition to those resulting from trivial invariance.

Abstract (translated by Google)
URL

http://arxiv.org/abs/1803.00885

PDF

http://arxiv.org/pdf/1803.00885


Similar Posts

Comments