Predicting Useful Neighborhoods for Lazy Local Learning
Authors: Aron Yu, Kristen Grauman
NeurIPS 2014 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the approach on image classification tasks on SUN and a Pascal and show its advantages over traditional global and local approaches. |
| Researcher Affiliation | Academia | Aron Yu University of Texas at Austin aron.yu@utexas.edu Kristen Grauman University of Texas at Austin grauman@cs.utexas.edu |
| Pseudocode | No | The paper describes its methods in narrative text rather than providing structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository for the described methodology. |
| Open Datasets | Yes | The SUN Attributes dataset [22] (SUN) contains 14,340 scene images labeled with binary attributes... The a Pascal training dataset [10] contains 6,440 object images labeled with attributes... |
| Dataset Splits | No | The paper states 'For each attribute, we compose a test set of 100 randomly chosen images (balanced between positives and negatives), and use all other images for T.' and 'All methods use the exact same image features and train linear SVMs, with the cost parameter cross-validated based on the Global baseline.' However, it does not explicitly define a separate 'validation' dataset split for hyperparameter tuning, beyond the implication of cross-validation for the cost parameter. |
| Hardware Specification | No | The paper mentions run-time figures and discusses computational overhead but does not provide specific hardware details such as GPU or CPU models used for the experiments. |
| Software Dependencies | No | The paper mentions using Matlab and the ITML metric learning algorithm, but does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | Yes | For each attribute, we compose a test set of 100 randomly chosen images (balanced between positives and negatives), and use all other images for T . This makes M = 14, 240 for SUN and M = 6, 340 for a Pascal. We use N = 2, 000 training neighborhoods for both, and set D = {2000, 1000} for SUN and a Pascal, roughly 15% of their original label indicator lengths. Generally higher values of D yield better accuracy (less compression), but for a greater expense. We fix the number of samples S = 100, and consider neighborhood sizes from k1 = 50 and k K = 500, in increments of 10 to 50. |