Stronger generalization bounds for deep nets via a compression approach
This provides theoretical justification for the empirical success of compressing deep nets, addressing a foundational issue in machine learning.
The paper tackles the problem of explaining why deep neural networks generalize well despite having many parameters, and shows generalization bounds that are orders of magnitude better in practice by using explicit and efficient compression techniques.
Deep nets generalize well despite having more parameters than the number of training samples. Recent works try to give an explanation using PAC-Bayes and Margin-based analyses, but do not as yet result in sample complexity bounds better than naive parameter counting. The current paper shows generalization bounds that're orders of magnitude better in practice. These rely upon new succinct reparametrizations of the trained net --- a compression that is explicit and efficient. These yield generalization bounds via a simple compression-based framework introduced here. Our results also provide some theoretical justification for widespread empirical success in compressing deep nets. Analysis of correctness of our compression relies upon some newly identified \textquotedblleft noise stability\textquotedblright properties of trained deep nets, which are also experimentally verified. The study of these properties and resulting generalization bounds are also extended to convolutional nets, which had eluded earlier attempts on proving generalization.