CaltechAUTHORS
  A Caltech Library Service

Deep regularization and direct training of the inner layers of Neural Networks with Kernel Flows

Yoo, Gene Ryan and Owhadi, Houman (2020) Deep regularization and direct training of the inner layers of Neural Networks with Kernel Flows. . (Unpublished) https://resolver.caltech.edu/CaltechAUTHORS:20201110-075343797

[img] PDF - Submitted Version
See Usage Policy.

1166Kb

Use this Persistent URL to link to this item: https://resolver.caltech.edu/CaltechAUTHORS:20201110-075343797

Abstract

We introduce a new regularization method for Artificial Neural Networks (ANNs) based on Kernel Flows (KFs). KFs were introduced as a method for kernel selection in regression/kriging based on the minimization of the loss of accuracy incurred by halving the number of interpolation points in random batches of the dataset. Writing f_θ(x) = (f^((n))_(θn)∘f^((n−1))_(θn−1)∘⋯∘f^((1))_(θ₁))(x) for the functional representation of compositional structure of the ANN, the inner layers outputs h^((i))(x) = (f^((i))_(θi)∘f^((i−1))_(θi−1)∘⋯∘f^((1))_(θ1))(x) define a hierarchy of feature maps and kernels k^((i))(x,x′) = exp(−γ_i∥h^((i))(x)−h^((i))(x′)∥²₂). When combined with a batch of the dataset these kernels produce KF losses e(i)₂ (the L² regression error incurred by using a random half of the batch to predict the other half) depending on parameters of inner layers θ₁,…,θ_i (and γ_i). The proposed method simply consists in aggregating a subset of these KF losses with a classical output loss. We test the proposed method on CNNs and WRNs without alteration of structure nor output classifier and report reduced test errors, decreased generalization gaps, and increased robustness to distribution shift without significant increase in computational complexity. We suspect that these results might be explained by the fact that while conventional training only employs a linear functional (a generalized moment) of the empirical distribution defined by the dataset and can be prone to trapping in the Neural Tangent Kernel regime (under over-parameterizations), the proposed loss function (defined as a nonlinear functional of the empirical distribution) effectively trains the underlying kernel defined by the CNN beyond regressing the data with that kernel.


Item Type:Report or Paper (Discussion Paper)
Related URLs:
URLURL TypeDescription
http://arxiv.org/abs/2002.08335arXivDiscussion Paper
ORCID:
AuthorORCID
Yoo, Gene Ryan0000-0002-5319-5599
Owhadi, Houman0000-0002-5677-1600
Additional Information:The authors gratefully acknowledge support by the Air Force Office of Scientific Research under award number FA9550-18-1-0271 (Games for Computation and Learning), Beyond Limits (Learning Optimal Models) and NASA/JPL (Earth 2050).
Funders:
Funding AgencyGrant Number
Air Force Office of Scientific Research (AFOSR)FA9550-18-1-0271
NASA/JPLEarth 2050
Record Number:CaltechAUTHORS:20201110-075343797
Persistent URL:https://resolver.caltech.edu/CaltechAUTHORS:20201110-075343797
Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:106580
Collection:CaltechAUTHORS
Deposited By: Tony Diaz
Deposited On:10 Nov 2020 16:16
Last Modified:30 Jan 2021 00:46

Repository Staff Only: item control page