Entropy-Based Logic Explanations of Neural Networks

Authors: Pietro Barbiero, Gabriele Ciravegna, Francesco Giannini, Pietro Lió, Marco Gori, Stefano Melacci6046-6054

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We consider four different case studies to demonstrate that: (i) this entropy-based criterion enables the distillation of concise logic explanations in safety-critical domains from clinical data to computer vision; (ii) the proposed approach outperforms state-of-the-art white-box models in terms of classification accuracy and matches black box performances. ... 5 Experiments The quality of the explanations and the classification performance of the proposed approach are quantitatively assessed and compared to state-of-the-art white-box models.
Researcher Affiliation Academia Pietro Barbiero1, Gabriele Ciravegna 2,3,4, Francesco Giannini 3, Pietro Li o 1, Marco Gori 3,4, Stefano Melacci 3, 1 University of Cambridge (UK) 2 Universit a di Firenze (Italy) 3 Universit a di Siena (Italy) 4 Universit e Cˆote d Azur (France)
Pseudocode No No section labeled 'Pseudocode' or 'Algorithm' or clearly formatted pseudocode block was found in the provided text.
Open Source Code Yes Finally, we share an implementation of the entropy layer, with extensive documentation and all the experiments in the public repository: https://github.com/pietrobarbiero/entropy-lens.
Open Datasets Yes Will we recover from ICU? (MIMIC-II). The Multiparameter Intelligent Monitoring in Intensive Care II (MIMICII, (Saeed et al. 2011; Goldberger et al. 2000)) is a public-access intensive care unit (ICU) database... What kind of democracy are we living in? (V-Dem). Varieties of Democracy (V-Dem, (Pemstein et al. 2018; Coppedge et al. 2021)) dataset contains a collection of indicators... What does parity mean? (MNIST Even/Odd). The Modified National Institute of Standards and Technology database (MNIST, (Le Cun 1998)) contains a large collection of images... What kind of bird is that? (CUB). The Caltech-UCSD Birds-200-2011 dataset (CUB, (Wah et al. 2011)) is a fine-grained classification dataset.
Dataset Splits Yes The results for each metric are reported in terms of mean and standard error, computed over a 5-fold cross validation (Krzywinski and Altman 2013).
Hardware Specification No No specific hardware details (e.g., CPU model, GPU model, memory specifications) used for running experiments were found in the provided text.
Software Dependencies No While software names like 'PyTorch' and 'Scikit-learn' are mentioned, specific version numbers for these or other dependencies were not provided in the text to ensure reproducibility.
Experiment Setup No No specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or detailed training configurations were found in the provided text.