Alternating Circulant Random Features for Semigroup Kernels

Authors: Yusuke Mukuta, Yoshitaka Ushiku, Tatsuya Harada

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

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments In this section, we experimentally evaluate the accuracy and computation time of the proposed method.
Researcher Affiliation Academia Yusuke Mukuta The University of Tokyo mukuta@mi.t.u-tokyo.ac.jp Yoshitaka Ushiku The University of Tokyo ushiku@mi.t.u-tokyo.ac.jp Tatsuya Harada The University of Tokyo and RIKEN AIP harada@mi.t.u-tokyo.ac.jp
Pseudocode No The paper describes the methods mathematically and textually but does not include structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of the proposed method.
Open Datasets Yes CUB-200 (Welinder et al. 2010) is a standard fine-grained object recognition dataset that consists of 200 bird species with 60 images per class. Stanford Dogs (Khosla et al. 2011) consists of approximately 20,000 images of 120 dog classes, and Caltech256 (Griffin, Holub, and Perona 2007) consists of approximately 30,600 images of 256 object classes.
Dataset Splits Yes We used the default train/test splits for the CUB-200 and Stanford Dogs datasets. For the Caltech256 dataset, we randomly sampled 25 images per class as training data and 30 images per class as test data.
Hardware Specification Yes We compared the computation time required to encode one input vector using an Intel(R) Xeon(R) CPU E52690 v3 @ 2.60GHz.
Software Dependencies No We implemented each method using Matlab with the -single Comp Thread option. We applied linear SVM implemented in LIBLINEAR (Fan et al. 2008) with C = 100. Specific version numbers for Matlab or LIBLINEAR are not provided.
Experiment Setup Yes As a kernel function, we used an exponential-semigroup kernel that was reported to exhibit good performance in (Yang et al. 2014), using the kernel parameter β = 0.1. We applied linear SVM implemented in LIBLINEAR (Fan et al. 2008) with C = 100.