LGJul 29, 2017
A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks
arXiv:1707.09564v2670 citations
AI Analysis
This provides theoretical guarantees for neural network generalization, which is incremental as it builds on existing PAC-Bayes and norm-based bound techniques.
The paper tackles the problem of deriving generalization bounds for feedforward neural networks, resulting in a bound expressed in terms of the product of spectral norms of layers and Frobenius norms of weights, using PAC-Bayes analysis.
We present a generalization bound for feedforward neural networks in terms of the product of the spectral norm of the layers and the Frobenius norm of the weights. The generalization bound is derived using a PAC-Bayes analysis.