Manifold Spanning Graphs

Authors: CJ Carey, Sridhar Mahadevan

AAAI 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The paper includes sections like 'Experimental Results', conducts evaluations on 'Parametric Swiss Roll' and 'MNIST digit clustering' datasets, and presents performance metrics in figures and tables (e.g., 'Figure 6: Summarized performance over 200 random swiss rolls', 'Table 1: Confusion matrices from the MNIST digit classification task.').
Researcher Affiliation Academia CJ Carey and Sridhar Mahadevan School of Computer Science University of Massachusetts, Amherst Amherst, Massachusetts, 01003 {ccarey,mahadeva}@cs.umass.edu
Pseudocode No The paper describes the algorithm's stages in prose (e.g., 'The high-level procedure can be divided into three stages: 1. Divide X into many small connected components...'), but does not include any formal pseudocode or algorithm blocks.
Open Source Code No Source code will be made available publicly on the author s website following publication.
Open Datasets Yes We consider the classic swiss roll dataset, shown in Figure 1. the MNIST dataset of 10, 000 handwritten digits (Le Cun and Cortes 1998)
Dataset Splits No The paper states it used '200 randomly-generated swiss rolls' and for MNIST, '30 unique images were selected at random from the 10, 000-image corpus to act as labeled examples' but does not specify explicit train/validation/test splits with percentages or counts for reproduction.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory used for running the experiments.
Software Dependencies No The paper states the algorithm was 'implemented in Python using open-source libraries (Oliphant 2007; Pedregosa et al. 2011)' but does not specify version numbers for Python or any of the libraries.
Experiment Setup Yes Inputs are X, an N D matrix of D-dimension points, d, an estimate of the intrinsic dimension of the underlying manifold, and m, the desired number of connected components in the final graph. Setting h d + 1 is a reasonable default, so this extra parameter requires no expert tuning. (Figure 7c caption) MSG: 1118/23042 4.9% bad edges (m = 10, d = 2)