Abstract:This paper introduces a new kernel-based classifier by viewing kernel matrices as generalized graphs and leveraging recent progress in graph embedding techniques. The proposed method facilitates fast and scalable kernel matrix embedding, and seamlessly integrates multiple kernels to enhance the learning process. Our theoretical analysis offers a population-level characterization of this approach using random variables. Empirically, our method demonstrates superior running time compared to standard approaches such as support vector machines and two-layer neural network, while achieving comparable classification accuracy across various simulated and real datasets.

What problem does this paper attempt to address?

### Problems the paper attempts to solve This paper aims to solve the problem of low computational efficiency of traditional kernel methods (such as Support Vector Machines, SVM) in multi - class classification tasks. Specifically: 1. **High computational complexity**: When dealing with large - scale data sets, traditional SVMs have high computational complexity because they need to calculate the complete kernel matrix and solve optimization problems. Especially in multi - class classification tasks, multiple binary classifiers need to be constructed, which further increases the computational burden. 2. **Limitations in kernel function selection**: SVMs are usually limited to using linear kernels or Gaussian kernels, and the selection of different kernel functions needs to be determined by cross - validation, which makes the process very time - consuming. 3. **Lack of scalability and flexibility**: Traditional methods are difficult to effectively integrate multiple kernel functions, thus limiting their flexibility and performance improvement in different scenarios. To solve these problems, the paper proposes a Multi - Kernel Encoder Classifier based on graph embedding. This method regards the kernel matrix as a generalized graph and uses graph embedding techniques to achieve fast and scalable kernel matrix embedding. This method can not only efficiently handle large - scale data sets, but also seamlessly integrate multiple kernel functions to enhance the learning process. ### Method overview 1. **Graph embedding perspective**: Regard the kernel matrix as the adjacency matrix of a weighted graph and use graph embedding techniques to convert it into a low - dimensional representation. 2. **Multi - kernel integration**: Classify the embedded features by Linear Discriminant Analysis (LDA) and select the optimal kernel function by cross - entropy. 3. **Theoretical analysis**: Theoretically analyze the method from a probabilistic perspective and prove that this method significantly improves computational efficiency while maintaining classification accuracy. ### Experimental results The experimental results show that this method performs well on a variety of simulated and real - world data sets. It is not only comparable to SVM in classification accuracy, but also has a significantly shorter running time. Especially when dealing with large - scale data, the advantages of this method are more obvious. ### Summary The method proposed in the paper solves the deficiencies of traditional kernel methods in computational efficiency and flexibility by introducing graph embedding techniques and multi - kernel integration, providing an efficient and flexible solution for large - scale data classification tasks.

Fast and Scalable Multi-Kernel Encoder Classifier

Multi-Kernel Transfer Extreme Learning Classification

Multiple-kernel-learning-based Extreme Learning Machine for Classification Design

Distance-Based Classifier Via the Kernel Trick.

An Improved Error-Correcting Output Coding Framework with Kernel-Based Decoding

Encoder Embedding for General Graph and Node Classification

Online Kernel-Based Structured Output SVM for Early Expression Detection

Fast Multi-View Segment Graph Kernel for Object Classification

Multiple Kernel Sparse Representation Based Classification

Multiple Locally Linear Kernel Machines

Two-stage sparse multi-kernel optimization classifier method for more accurate and explainable prediction

How to Scale Up Kernel Methods to Be As Good As Deep Neural Nets

Large Scale Online Kernel Classification

High-performance Kernel Machines with Implicit Distributed Optimization and Randomization

Simple, Fast, and Flexible Framework for Matrix Completion with Infinite Width Neural Networks

AKBR: Learning Adaptive Kernel-based Representations for Graph Classification

A New Convolutional Kernel Classifier for Hyperspectral Image Classification

A Deep Learning Approach To Multiple Kernel Fusion

Accurate, Efficient and Scalable Graph Embedding

MK-MSMCR: An efficient multiple kernel approach to multi-class classification