Randomized Kernel Selection With Spectra of Multilevel Circulant Matrices

Authors: Lizhong Ding, Shizhong Liao, Yong Liu, Peng Yang, Xin Gao

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that our randomized kernel selection criteria are significantly more efficient than the existing classic and widely-used criteria while preserving similar predictive performance.
Researcher Affiliation Academia 1 King Abdullah University of Science and Technology (KAUST), Computational Bioscience Research Center (CBRC), Computer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division, Thuwal, 23955-6900, Saudi Arabia 2 School of Computer Science and Technology, Tianjin University, Tianjin 300350, China 3 Institute of Information Engineering, CAS, Beijing, China
Pseudocode No The paper describes its methods through mathematical formulations and prose, but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statement about releasing source code, nor does it provide a link to a code repository.
Open Datasets Yes A variety of datasets that cover the number of data points ranging from 7200 up to more than 245000 and the number of features ranging from 3 up to 47,236 are selected from the UCI dataset repository3, the LIBSVM dataset repository4, and the KEEL dataset repository5.
Dataset Splits No We randomly partition each dataset into two parts, with 50% of the data randomly chosen for training and the rest reserved for testing.
Hardware Specification No It is worth noting that this experiment is conducted on a PC, while the first experiment is conducted on our computing cluster.
Software Dependencies No We adopt least square support vector machine (LSSVM) as the learning algorithm.
Experiment Setup Yes The set of Gaussian kernels κ(x x ) = exp( γ x x 2 2) with a variable bandwidth parameter γ {2i, i = 8, 7, . . . , 5, 6} is adopted as the candidate kernel set. The parameter D in the randomized criteria is an important parameter both for the approximation quality and computational efficiency (O(Πm ln(Πm)) = O(l D ln(l D))). We conduct experiments for different values of D (50, 100, 200, 500, 1000, 2000). Finally, we fix D = 100... Since the focus of this work is not on tuning the regularization parameter, it is set as a fixed value 1.