Kernel-based Translations of Convolutional Networks
This work bridges a gap in machine learning theory for researchers, though it is incremental as it translates existing architectures rather than introducing new capabilities.
The paper tackles the perceived difference between convolutional neural networks and kernel methods by providing a systematic translation of ConvNets into Convolutional Kernel Networks, showing that CKNs perform on par with their ConvNet counterparts in experiments.
Convolutional Neural Networks, as most artificial neural networks, are commonly viewed as methods different in essence from kernel-based methods. We provide a systematic translation of Convolutional Neural Networks (ConvNets) into their kernel-based counterparts, Convolutional Kernel Networks (CKNs), and demonstrate that this perception is unfounded both formally and empirically. We show that, given a Convolutional Neural Network, we can design a corresponding Convolutional Kernel Network, easily trainable using a new stochastic gradient algorithm based on an accurate gradient computation, that performs on par with its Convolutional Neural Network counterpart. We present experimental results supporting our claims on landmark ConvNet architectures comparing each ConvNet to its CKN counterpart over several parameter settings.