Essentially no barriers in neural network energy landscape
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 CIFA...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article (Journal) Chapter/Article |
| Language: | English |
| Published: |
2 Mar 2018
|
| In: |
Arxiv
|
| Online Access: | kostenfrei
|
| Author Notes: | Felix Draxler, Kambis Veschgini, Manfred Salmhofer, Fred A. Hamprecht |
| Summary: | 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. |
|---|---|
| Item Description: | Identifizierung der Ressource nach: Last revised 22 Feb 2019 Gesehen am 15.12.2020 |
| Physical Description: | Online Resource |