Taming Discrete Integration via the Boon of Dimensionality
Authors: Jeffrey Dudek, Dror Fried, Kuldeep S Meel
NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We augment our proposed reduction, called De Weight, with a state of the art efficient approximate model counter and perform detailed empirical analysis over benchmarks arising from neural network verification domains, an emerging application area of critical importance. De Weight, to the best of our knowledge, is the first technique to compute estimates with provable guarantees for this class of benchmarks. |
| Researcher Affiliation | Academia | Jeffrey M. Dudek Rice University Houston, USA; Dror Fried The Open University of Israel Israel; Kuldeep S. Meel National University of Singapore Singapore |
| Pseudocode | Yes | Algorithm 1 An algorithm for finding a nearest m-bit fractions to p/q. Input: Fractions (a1, b1), (p, q), (a2, b2), non-negative integer m Output: Fractions (c1, d1) |
| Open Source Code | Yes | All code is available in a public repository at https://github.com/meelgroup/deweight |
| Open Datasets | Yes | Formulas for robustness and trojan attack effectiveness are based on a binarized neural network trained on the MNIST [31] dataset. ... Formulas for fairness are based on a binarized neural network trained on the UCI Adult dataset [3]. |
| Dataset Splits | No | The paper uses established benchmarks for evaluation but does not specify how the data for these benchmarks was split into training, validation, or testing sets for the discrete integration experiments conducted in this paper. It mentions a "sampling set" but not explicit data splits for the experiments. |
| Hardware Specification | Yes | All experiments are performed on 2.5 GHz CPUs with 24 cores and 96GB RAM. |
| Software Dependencies | No | The paper mentions using "Approx MC4 [43]" but does not specify a version number for this or any other software dependency. |
| Experiment Setup | Yes | In keeping in line with the prior studies, we configure a tolerable error ϵ = 0.8 and confidence parameter δ = 0.2 as defaults throughout the evaluation. |