Improving Sample Efficiency with Normalized RBF Kernels

TL;DR

This paper demonstrates that neural networks with normalized RBF kernels as output layers improve sample efficiency and class separation, especially on CIFAR datasets, by using a novel two-phase training method.

Contribution

It introduces a two-phase training approach for neural networks with normalized RBF kernels, enhancing sample efficiency and class separation compared to SoftMax layers.

Findings

Higher sample efficiency on CIFAR-10 and CIFAR-100

Improved class compactness and separability

Effective two-phase training method

Abstract

In deep learning models, learning more with less data is becoming more important. This paper explores how neural networks with normalized Radial Basis Function (RBF) kernels can be trained to achieve better sample efficiency. Moreover, we show how this kind of output layer can find embedding spaces where the classes are compact and well-separated. In order to achieve this, we propose a two-phase method to train those type of neural networks on classification tasks. Experiments on CIFAR-10 and CIFAR-100 show that networks with normalized kernels as output layer can achieve higher sample efficiency, high compactness and well-separability through the presented method in comparison to networks with SoftMax output layer.