AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Entropy-Based Logic Explanations of Neural Networks

Pietro Barbiero, Gabriele Ciravegna, Francesco Giannini, Pietro Lió, Marco Gori, Stefano Melacci

AAAI 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Post-hoc explanation analysis · 65 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use entropy-based concept selection and logic explanations instead of a decision treeMIMIC-II classification · MIMIC-IIaccuracy (%): 77.53 ± 1.45 -> 79.05 ± 1.35Improved
Use entropy-based concept selection and logic explanations instead of BRLV-Dem classification · V-Demaccuracy (%): 91.23 ± 0.75 -> 94.51 ± 0.48Improved
Use entropy-based concept selection and logic explanations instead of a standard neural networkV-Dem classification · V-Demaccuracy (%): 94.53 ± 1.17 -> 94.51 ± 0.48No clear effect
Use entropy-based concept selection and logic explanations instead of a random forestMNIST even/odd classification · MNISTaccuracy (%): 99.96 ± 0.01 -> 99.81 ± 0.02Hurt
Use entropy-based concept selection and logic explanations instead of BRLCUB classification · CUBaccuracy (%): 90.79 ± 0.34 -> 92.95 ± 0.20Improved
Use entropy-based explanations instead of ψ net explanationsExplanation fidelity on all four datasets · MIMIC-II, V-Dem, MNIST, CUBfidelity (%): MIMIC-II: 51.63 ± 6.67 -> 79.11 ± 2.02; V-Dem: 69.67 ± 10.43 -> 90.90 ± 1.23; MNIST: 65.68 ± 5.05 -> 99.63 ± 0.00; CUB: Improved
Replace L1 regularization with entropy regularization while keeping the Entropy Layer (ablation)Toy Boolean classification tasks · Toy datasetconcept relevance identificationImproved
Add the Entropy Layer as the first layer instead of using a standard linear first layer (ablation)Toy Boolean classification tasks · Toy datasetconcept relevance interpretabilityImproved
Vary τ and λ across the reported V-Dem grid (ablation)Entropy net · V-Demmodel accuracy (%): 94.05 ± 0.58 to 94.86 ± 0.23No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy