A Deep Learning Approach To Multiple Kernel Fusion

TL;DR

This paper introduces a deep learning method for multiple kernel fusion that creates dense data embeddings and uses a neural network with kernel dropout, achieving state-of-the-art results in activity recognition.

Contribution

It proposes a novel deep neural network architecture with kernel dropout for effective multiple kernel fusion, replacing traditional optimization-based MKL methods.

Findings

Achieves state-of-the-art performance on activity recognition dataset

Effective kernel fusion through dense embeddings and neural network architecture

Kernel dropout improves model robustness and fusion quality

Abstract

Kernel fusion is a popular and effective approach for combining multiple features that characterize different aspects of data. Traditional approaches for Multiple Kernel Learning (MKL) attempt to learn the parameters for combining the kernels through sophisticated optimization procedures. In this paper, we propose an alternative approach that creates dense embeddings for data using the kernel similarities and adopts a deep neural network architecture for fusing the embeddings. In order to improve the effectiveness of this network, we introduce the kernel dropout regularization strategy coupled with the use of an expanded set of composition kernels. Experiment results on a real-world activity recognition dataset show that the proposed architecture is effective in fusing kernels and achieves state-of-the-art performance.