Diving into the shallows: a computational perspective on large-scale shallow learning

Authors: SIYUAN MA, Mikhail Belkin

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 6 Experimental Results
Researcher Affiliation Academia Siyuan Ma Mikhail Belkin Department of Computer Science and Engineering The Ohio State University {masi, mbelkin}@cse.ohio-state.edu
Pseudocode Yes Algorithm: Eigen Pro(X, y, k, m, η, τ, M)
Open Source Code Yes In the second part of the paper we propose Eigen Pro iteration (see http://www.github.com/Eigen Pro for the code)
Open Datasets Yes Dataset Size Gaussian Laplace Cauchy Eig Pro Pega Eig Pro Pega Eig Pro Pega MNIST 6 104 ... CIFAR-10 5 104 ... SVHN 7 104 ... HINT-S 5 104 ... TIMIT 1 106 ... SUSY 4 106
Dataset Splits No The paper mentions "train" and "test" data in tables but does not explicitly provide information on validation splits or methodology.
Hardware Specification Yes Experiments were run on a workstation with 128GB main memory, two Intel Xeon(R) E5-2620 CPUs, and one GTX Titan X (Maxwell) GPU.
Software Dependencies No The paper mentions software like Pegasos and Random Fourier Features but does not specify their version numbers or other ancillary software dependencies with versions.
Experiment Setup Yes For consistent comparison, all iterative methods use mini-batch of size m = 256. Eigen Pro preconditioner is constructed using the top k = 160 eigenvectors of a subsampled dataset of size M = 4800. For Eigen Pro-RF, we set the damping factor τ = 1/4. For primal Eigen Pro τ = 1.