This is a continuation of Julien Mairal‘s guest post on CNNs, see part I here.

**Stability to deformations of convolutional neural networks**

In their ICML paper Zhang et al. introduce a functional space for CNNs with one layer, by noticing that for some dot-product kernels, smoothed variants of rectified linear unit activation functions (ReLU) live in the corresponding RKHS, see also this paper and that one. By following a similar reasoning with multiple layers, it is then possible to show that the functional space described in part I contains CNNs with such smoothed ReLU, and that the norm of such networks can be controlled by the spectral norms of filter matrices. This is consistent with previous measures of complexity for CNNs, see this paper by Bartlett et al.

A perhaps more interesting finding is that the abstract representation , which only depends on the network architecture, may provide near-translation invariance and stability to small image deformations while preserving information—that is, can be recovered from . The original characterization we use was introduced by Mallat in his paper on the scattering transform—a multilayer architecture akin to CNNs based on wavelets, and was extended to by Alberto Bietti, who should be credited for all the hard work here.

Our goal is to understand under which conditions it is possible to obtain a representation that (i) is near-translation invariant, (ii) is stable to deformations, (iii) preserves signal information. Given a -diffeomorphism and denoting by its action operator (for an image defined on the continuous domain ), the main stability bound we obtain is the following one, see Theorem 7 in Mallat’s paper if , for all ,

where are universal constants, is the scale parameter of the pooling operator corresponding to the “amount of pooling” performed up to the last layer, is the maximum pixel displacement and represents the maximum amount of deformation, see the paper for the precise definitions of all these quantities. Note that when , the representation becomes translation invariant: indeed, consider the particular case of being a translation, then and .

The stability bound and a few additional results tell us a few things about the network architecture: (a) small patches lead to more stable representations (the dependency is hidden in ); (b) signal preservation for discrete signals requires small subsampling factors (and thus small pooling) between layers. In such a setting, the scale parameter still grows exponentially with and near translation invariance may be achieved with several layers.

Interestingly, we may now come back to the Cauchy-Schwarz inequality from part 1, and note that if is stable, the RKHS norm is then a natural quantity that provides stability to deformations to the prediction function , in addition to measuring model complexity in a traditional sense.

**Feature learning in RKHSs and convolutional kernel networks**

The previous paragraph is devoted to the characterization of convolutional architectures such as CNNs but the previous kernel construction can in fact be used to derive more traditional kernel methods. After all, why should one spend efforts defining a kernel between images if not to use it?

This can be achieved by considering finite-dimensional approximations of the previous feature maps. In order to shorten the presentation, we simply describe the main idea based on the Nystrom approximation and refer to the paper for more details. Approximating the infinite-dimensional feature maps (see the figure at the top of part I) can be done by projecting each point in onto a -dimensional subspace leading to a finite-dimensional feature map akin to CNNs, see the figure at the top of the post.

By parametrizing with anchor points , and using a dot-product kernel, a patch from is encoded through the mapping function

where is applied pointwise. Then, computing from admits a CNN interpretation, where only the normalization and the matrix multiplication by are not standard operations. It remains now to choose the anchor points:

**kernel approximation:**a first approach consists of using a variant of the Nystrom method, see this paper and that one. When plugging the corresponding image representation in a linear classifier, the resulting approach behaves as a classical kernel machine. Empirically, we observe that the higher the number of anchor points, the better the kernel approximation, and the higher the accuracy. For instance, a two-layer network with a -dimensional representations achieves about accuracy on CIFAR-10 without data augmentation (see here).**back-propagation, feature selection**: learning the anchor points can also be done as in a traditional CNN, by optimizing them end-to-end. This allows using deeper lower-dimensional architectures and empirically seems to perform better when enough data is available, e.g., accuracy on CIFAR-10 with simple data augmentation. There, the subspaces are not learned anymore to provide the best kernel approximation, but the model seems to perform a sort of feature selection in each layer’s RKHS , which is not well understood yet (This feature selection interpretation is due to my collaborator Laurent Jacob).

Note that the first CKN model published here was based on a different approximation principle, which was not compatible with end-to-end training. We found this to be less scalable and effective.

**Other links between neural networks and kernel methods**

Finally, other links between kernels and infinitely-wide neural networks with random weights are classical, but they were not the topic of this blog post (they should be the topic of another one!). In a nutshell, for a large collection of weights distributions and nonlinear functions , the following quantity admits an analytical form

where the terms may be seen as an infinitely-wide single-layer neural network. The first time such a relation appears is likely to be in the PhD thesis of Radford Neal with a Gaussian process interpretation, and it was revisited later by Le Roux and Bengio and by Cho and Saul with multilayer models.

In particular, when is the rectified linear unit and follows a Gaussian distribution, it is known that we recover the arc-cosine kernel. We may also note that random Fourier features also yield a similar interpretation.

Other important links have also been drawn recently between kernel regression and strongly over-parametrized neural networks, see this paper and that one, which is another exciting story.

## By kons iş ilanları October 13, 2019 - 9:05 am

thank you https://www.solistkonsilanlari.com

## By ankara ikinci el eşya October 12, 2019 - 11:40 am

Thanks you

## By Beysülen Teknik October 1, 2019 - 6:02 am

İzmir evka2 de klima servisi olarak klima, beyaz eşya ve kombi servis hizmet veriyoruz.