Using Stratified Sampling to Improve LIME Image Explanations
Authors: Muhammad Rashid, Elvio G. Amparore, Enrico Ferrari, Damiano Verda
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show the efficacy of the proposed approach. and provide empirical proofs of the advantage of using stratified sampling for LIME Image on a popular dataset. |
| Researcher Affiliation | Collaboration | 1University of Torino, Computer Science Department, C.so Svizzera 185, 10149 Torino, Italy 2Rulex Innovation Labs, Via Felice Romani 9, 16122 Genova, Italy |
| Pseudocode | Yes | Algorithm 1: Neighborhood sampling strategies |
| Open Source Code | Yes | The LIME Image with stratified sampling is available at: https://github.com/rashidrao-pk/lime stratified |
| Open Datasets | Yes | To better quantify the effect, we took the first 150 images of the Image Net Object Localization dataset (Addison Howard 2018). |
| Dataset Splits | No | The paper states it uses 'the first 150 images of the Image Net Object Localization dataset' but does not specify any train/validation/test splits for these images in the context of their experiments. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types with speeds, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | Yes | All code needed to replicate the experiments (including the requirements.txt with the library versions used) can be found at: https://github.com/rashidrao-pk/lime-stratified-examples |
| Experiment Setup | Yes | For each image we performed a dichotomic search on the max dist hyperparameter to find a configuration of quick shift that results in a number of superpixels k equal to 50, 100, 150 and 200. For each range, we run 10 times LIME Image with both the Monte Carlo and the stratified sampling using n=1000 samples....Using: kernel size = 4, max dist = 7, ratio = 0.2. and σ = 0.25 (by default) is the kernel width. |