Testing Sparsity over Known and Unknown Bases

Authors: Siddharth Barman, Arnab Bhattacharyya, Suprovat Ghoshal

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

Reproducibility Variable Result LLM Response
Research Type Experimental This section provides experimental evidence which supplements our theoretical results. For the empirical study, we use the classic Barbara image (which is of size 512 512 pixels). Specifically, we consider 9 subimages of size 100 100 pixels each (see Figure 1). For each such sub-image, we compute a matrix representation (by the standard technique of subdividing the images into patches, see, e.g., (Elad & Aharon, 2006)). In particular, each sub-image is represented as a matrix Y of dimension 64 8649. Then, for each matrix Y corresponding to a sub-image, we estimate the gaussian width of the ℓ2-column normalized matrix. In addition, setting the number of atoms m = 100 and sparsity k = 10, we run the k-SVD algorithm for 50 iterations and record the reconstruction error.
Researcher Affiliation Academia Department of Computer Science and Automation, Indian Institute of Science, Bangalore, India. Correspondence to: Siddharth Barman <barman@iisc.ac.in>, Arnab Bhattacharyya <arnabb@iisc.ac.in>, Suprovat Ghoshal <suprovat@iisc.ac.in>.
Pseudocode Yes Algorithm 1 Sparse Test Unknown; Algorithm 2 Sparse Test-Known Design
Open Source Code No The paper does not provide any specific links or statements about the availability of its source code.
Open Datasets Yes For the empirical study, we use the classic Barbara image (which is of size 512 512 pixels). Specifically, we consider 9 subimages of size 100 100 pixels each (see Figure 1).
Dataset Splits No The paper mentions using sub-images of the Barbara image for empirical study but does not specify any training, validation, or test splits for this data. It describes computing representations and estimating gaussian width, not model training with explicit data splits.
Hardware Specification No The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory, or specific computer specifications) used to run its experiments.
Software Dependencies No The paper mentions running the 'k-SVD algorithm' but does not specify the version of this algorithm or any other software dependencies (libraries, frameworks, etc.) with specific version numbers.
Experiment Setup Yes In addition, setting the number of atoms m = 100 and sparsity k = 10, we run the k-SVD algorithm for 50 iterations and record the reconstruction error.