Heating up decision boundaries: isocapacitory saturation, adversarial scenarios and generalization bounds

Authors: Bogdan Georgiev, Lukas Franken, Mayukh Mukherjee

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimentally, we observe a certain capacitory trend over different adversarial defense strategies...We study the capacitory metric on the well-known CIFAR10 and MNIST datasets...In Fig. 4 we collect the statistics of the WRN and Le Net models on CIFAR10 and MNIST, respectively.
Researcher Affiliation Collaboration Bogdan Georgiev Fraunhofer IAIS, ML2R bogdan.m.georgiev@gmail.com Lukas Franken Fraunhofer IAIS, ML2R, University of Cologne lukas.b.franken@gmail.com Mayukh Mukherjee IIT Bombay mathmukherjee@gmail.com
Pseudocode Yes Algorithm 1: Compressing a matrix A Rh1 h2
Open Source Code Yes We provide the training as well as the sampling code for our experiments.
Open Datasets Yes We study the capacitory metric on the well-known CIFAR10 and MNIST datasets
Dataset Splits Yes The MNIST is a 784-dimensional dataset that consists of 60000 images of handwritten digits whose dimensions are (28, 28); 50000 images were used for training and 10000 for validation. CIFAR-10 is collection of 60000 32-by-32 color images (i.e. a 3072-dimensional dataset) corresponding to 10 different classes...; 50000 images were used for training and 10000 for validation.
Hardware Specification Yes The experimental section of the work was conducted mainly on a CUDA 10.2 GPU-rack consisting of four NVIDA TITAN V units: this includes the model training as well as Brownian motion sampling and further statistics.
Software Dependencies Yes The neural network framework of choice was Py Torch 1.5.
Experiment Setup Yes The training of the Wide-Res Nets followed the framework provided by Cubuk et al. (2018) with weight decay 5e-4, batch size 128 and a decrease of the initial learning rate of 0.1 by a factor 0.2 at epochs 60, 120 and 160. The Res Nets were trained with weight decay 1e-4 respectively and step wise decrease of the learning rate 0.1 by a factor 0.1 at epochs 100 and 150. ... We trained a Le Net-5 architecture Le Cun et al. (1998) over 50 epochs with a learning rate 1e-3 and weight decay 5e-4, batch size of 64, while optimizing cross entropy loss using root mean square propagation.