Analyzing Stability of Convolutional Neural Networks in the Frequency Domain

11/10/2015
by   Elnaz J. Heravi, et al.
0

Understanding the internal process of ConvNets is commonly done using visualization techniques. However, these techniques do not usually provide a tool for estimating the stability of a ConvNet against noise. In this paper, we show how to analyze a ConvNet in the frequency domain using a 4-dimensional visualization technique. Using the frequency domain analysis, we show the reason that a ConvNet might be sensitive to a very low magnitude additive noise. Our experiments on a few ConvNets trained on different datasets revealed that convolution kernels of a trained ConvNet usually pass most of the frequencies and they are not able to effectively eliminate the effect of high frequencies. Our next experiments shows that a convolution kernel which has a more concentrated frequency response could be more stable. Finally, we show that fine-tuning a ConvNet using a training set augmented with noisy images can produce more stable ConvNets.

READ FULL TEXT

page 3

page 4

page 8

research
10/03/2022

Convolutional networks inherit frequency sensitivity from image statistics

It is widely acknowledged that trained convolutional neural networks (CN...
research
11/16/2021

Learning with convolution and pooling operations in kernel methods

Recent empirical work has shown that hierarchical convolutional kernels ...
research
10/28/2020

Ground Roll Suppression using Convolutional Neural Networks

Seismic data processing plays a major role in seismic exploration as it ...
research
09/16/2021

Dense Pruning of Pointwise Convolutions in the Frequency Domain

Depthwise separable convolutions and frequency-domain convolutions are t...
research
05/04/2021

Semiparametric Spatiotemporal Model with Mixed Frequencies

In modelling time series data coming from different sources, frequencies...
research
03/16/2022

Understanding robustness and generalization of artificial neural networks through Fourier masks

Despite the enormous success of artificial neural networks (ANNs) in man...
research
04/19/2021

SAPE: Spatially-Adaptive Progressive Encoding for Neural Optimization

Multilayer-perceptrons (MLP) are known to struggle with learning functio...

Please sign up or login with your details

Forgot password? Click here to reset