Feature and Decision Level Fusion Using Multiple Kernel Learning and Fuzzy Integrals

Pinar, Anthony J.

doi:10.37099/mtu.dc.etdr/375

Cited by 1 publication

(1 citation statement)

References 62 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Based on the normalized f k values and our respective target labels, formulate and solve a quadratic programming problem (see [24]) to obtain the 2 m ´2 free Choquet integral parameters (g). As Algorithms 2 and 3 show, DeFIMKLp is based on: (1) running a different decision maker (e.g., SVM) for each kernel; (2) normalizing those decision makers outputs; (3) forming a quadratic optimization problem using the normalized outputs and known labels to learn the Choquet integral parameters g; and then (4) for testing, run the kernel machines, normalize their outputs and do nonlinear aggregation with the Choquet integral using the learned g. Note, in [11] we showed how to do DeFIMKLp, GAMKLp, and MKLGLp for large numbers of samples via Nystrom kernel sampling and linearization.…”

Section: Algorithm 2: Defimkl Classifier Trainingmentioning

confidence: 99%

Kernel Matrix-Based Heuristic Multiple Kernel Learning

2022

View full text Add to dashboard Cite

Kernel theory is a demonstrated tool that has made its way into nearly all areas of machine learning. However, a serious limitation of kernel methods is knowing which kernel is needed in practice. Multiple kernel learning (MKL) is an attempt to learn a new tailored kernel through the aggregation of a set of valid known kernels. There are generally three approaches to MKL: fixed rules, heuristics, and optimization. Optimization is the most popular; however, a shortcoming of most optimization approaches is that they are tightly coupled with the underlying objective function and overfitting occurs. Herein, we take a different approach to MKL. Specifically, we explore different divergence measures on the values in the kernel matrices and in the reproducing kernel Hilbert space (RKHS). Experiments on benchmark datasets and a computer vision feature learning task in explosive hazard detection demonstrate the effectiveness and generalizability of our proposed methods.

show abstract

Section: Algorithm 2: Defimkl Classifier Trainingmentioning

confidence: 99%