Less is More: Nyström Computational Regularization

Authors: Alessandro Rudi, Raffaello Camoriano, Lorenzo Rosasco

NeurIPS 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental analysis shows that the considered approach achieves state of the art performances on benchmark large scale datasets.
Researcher Affiliation Academia Universit a degli Studi di Genova DIBRIS, Via Dodecaneso 35, Genova, Italy Istituto Italiano di Tecnologia i Cub Facility, Via Morego 30, Genova, Italy Massachusetts Institute of Technology and Istituto Italiano di Tecnologia Laboratory for Computational and Statistical Learning, Cambridge, MA 02139, USA
Pseudocode Yes Algorithm 1: Incremental Nystr om KRLS.
Open Source Code Yes 2The code for Algorithm 1 is available at lcsl.github.io/Nystrom Co Re.
Open Datasets Yes We consider the pumadyn32nh (n = 8192, d = 32), the breast cancer (n = 569, d = 30), and the cpu Small (n = 8192, d = 12) datasets4. 4www.cs.toronto.edu/ delve and archive.ics.uci.edu/ml/datasets
Dataset Splits Yes We randomly split the training part in a training set and a validation set (80% and 20% of the n training points, respectively) for parameter tuning via cross-validation.
Hardware Specification Yes The model selection times, measured on a server with 12 2.10GHz Intel Xeon E5-2620 v2 CPUs and 132 GB of RAM, are reported in Figure 2.
Software Dependencies No The paper mentions 'Cholesky rank-one update formulas' and 'linear algebra libraries' but does not provide specific software names with version numbers for dependencies.
Experiment Setup Yes We empirically study the properties of Algorithm 1, considering a Gaussian kernel of width σ. The λ values are logarithmically spaced, while the m values are linearly spaced. The ranges and kernel bandwidths, chosen according to preliminary tests on the data, are σ = 2.66, λ ∈ [10−7, 1], m ∈ [10, 1000] for pumadyn32nh, σ = 0.9, λ ∈ [10−12, 10−3], m ∈ [5, 300] for breast cancer, and σ = 0.1, λ ∈ [10−15, 10−12], m ∈ [100, 5000] for cpu Small.