Explaining A Black-box By Using A Deep Variational Information Bottleneck Approach
Authors: Seojin Bang, Pengtao Xie, Heewook Lee, Wei Wu, Eric Xing11396-11404
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate VIBI on three datasets and compare with state-of-the-art interpretable machine learning methods in terms of both interpretability and fidelity evaluated by human and quantitative metrics. |
| Researcher Affiliation | Collaboration | 1Carnegie Mellon University, PA, USA 2Petuum Inc., PA, USA 3University of California San Diego, CA, USA 4Arizona State University, AZ, USA |
| Pseudocode | No | The paper describes the proposed approach using mathematical formulations and descriptive text, but it does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The code is publicly available on Git Hub: github.com/Seojin Bang/VIBI. |
| Open Datasets | Yes | The IMDB (Maas et al. 2011) is a large text dataset containing movie reviews labeled by sentiment (positive/negative). ... The MNIST (Le Cun et al. 1998) is a large dataset contains 28 28 sized images of handwritten digits (0 to 9). ... These approaches rely on known interacting TCR-epitope pairs available from VDJdb (Shugay et al. 2017) and IEDB (Vita et al. 2014) |
| Dataset Splits | Yes | We grouped the reviews into training, validation, and test sets, which have 25,000, 12,500, and 12,500 reviews respectively. ... We grouped the images into training, validation, and test sets, which have 50,000, 10,000, and 10,000 images respectively |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running the experiments, such as GPU/CPU models, memory, or cloud instance types. |
| Software Dependencies | No | The paper mentions algorithms like Adam and techniques like Gumbel-softmax, and various software methods (LIME, SHAP, Saliency Map), but it does not specify software dependencies with version numbers (e.g., 'PyTorch 1.9' or 'TensorFlow 2.x'). |
| Experiment Setup | Yes | The settings of hyperparameter tuning include (bold indicate the choice for our final model): the temperature for Gumbel-softmax approximation τ {0.1, 0.2, 0.5, 0.7, 1}, learning rate 5 10 3, 10 3, 5 10 4, 10 4, 5 10 5} and β {0, 0.001, 0.01, 0.1, 1, 10, 100}. We use Adam algorithm (Kingma and Ba 2014) with batch size 100 for MNIST and 50 for IMDB, the coefficients used for computing running averages of gradient and its square (β1, β2) = (0.5, 0.999), and ϵ = 10 8. |