A Geometric Approach to Archetypal Analysis via Sparse Projections

Authors: Vinayak Abrol, Pulkit Sharma

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that GAA is considerably faster while performing comparable to existing methods for tasks such as classification, data visualization/categorization.
Researcher Affiliation Academia 1Mathematical Institute, University of Oxford. 2Department of Engineering Science, University of Oxford.
Pseudocode Yes Algorithm 1 Greedy Archetypal Analysis (GAA) algorithm
Open Source Code No The paper mentions a third-party tool 'SPAMS toolbox' with a link (http://spams-devel.gforge.inria.fr/), but it does not state that the authors' own implementation code for GAA is open-source or publicly available.
Open Datasets Yes This experiment uses the kernel AA model for digit classification task on USPS dataset (Bache & Lichman, 2013), to see if similar performance could be achieved by using AA instead of DL. In this experiment, we analyze the SUN attribute image dataset: a subset of SUN image database (Xiao et al., 2010).
Dataset Splits No The paper states it learns dictionaries with specific parameters for different algorithms, such as '300 atoms/archetypes, cardinality in SC step 5, polynomial kernel of degree 4, error tolerance ϵ = 10 4 and maximum iterations 200'. However, it does not provide specific details on how the dataset was split into training, validation, and test sets (e.g., percentages or sample counts), nor does it reference a standard split methodology with citations.
Hardware Specification Yes The empirical computational times are measured on a Quad-Core Intel i7 machine at 3.5 GHz, 12 GB RAM, using MATLAB and under Windows10 operating system.
Software Dependencies No The paper mentions 'GAA employ the active-set solver for QP from SPAMS toolbox' and that experiments were conducted 'using MATLAB and under Windows10 operating system'. However, it does not provide specific version numbers for MATLAB, Windows10, or the SPAMS toolbox, nor for any other programming languages or libraries used in their implementation.
Experiment Setup Yes Specifically, dictionary for each class using kernel-GAA (KGAA), kernel-AAAS (KAAAS), kernel-KSVD (KKSVD) (Van Nguyen et al., 2013) and kernel sparse greedy dictionary (KSGD) (Abrol et al., 2016) algorithm is learned with the following parameters: 300 atoms/archetypes, cardinality in SC step 5, polynomial kernel of degree 4, error tolerance ϵ = 10 4 and maximum iterations 200.