%0 Journal Article %1 sun2019lightlike %A Sun, Ke %A Nielsen, Frank %D 2019 %K deep-learning manifolds %T Lightlike Neuromanifolds, Occam's Razor and Deep Learning %U http://arxiv.org/abs/1905.11027 %X Why do deep neural networks generalize with a very high dimensional parameter space? We took an information theoretic approach. We find that the dimensionality of the parameter space can be studied by singular semi-Riemannian geometry and is upper-bounded by the sample size. We adapt Fisher information to this singular neuromanifold. We use random matrix theory to derive a minimum description length of a deep learning model, where the spectrum of the Fisher information matrix plays a key role to improve generalisation.

@article{sun2019lightlike, abstract = {Why do deep neural networks generalize with a very high dimensional parameter space? We took an information theoretic approach. We find that the dimensionality of the parameter space can be studied by singular semi-Riemannian geometry and is upper-bounded by the sample size. We adapt Fisher information to this singular neuromanifold. We use random matrix theory to derive a minimum description length of a deep learning model, where the spectrum of the Fisher information matrix plays a key role to improve generalisation.}, added-at = {2019-12-11T14:18:43.000+0100}, author = {Sun, Ke and Nielsen, Frank}, biburl = {https://www.bibsonomy.org/bibtex/2a61aabc7a5efc55981b77a3c80442f6c/kirk86}, description = {[1905.11027] Lightlike Neuromanifolds, Occam's Razor and Deep Learning}, interhash = {2d356753d24fb32e305a019b96240565}, intrahash = {a61aabc7a5efc55981b77a3c80442f6c}, keywords = {deep-learning manifolds}, note = {cite arxiv:1905.11027Comment: Submitted to NeurIPS 2019}, timestamp = {2019-12-11T14:18:43.000+0100}, title = {Lightlike Neuromanifolds, Occam's Razor and Deep Learning}, url = {http://arxiv.org/abs/1905.11027}, year = 2019 }