Adversarial Risk via Optimal Transport and Optimal Couplings
Authors: Muni Sreenivas Pydi, Varun Jog
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we present lower bounds on the optimal adversarial risk for empirical distributions derived from several real world datasets. Using this methodology, we evaluate the optimal risk for 2 and adversaries for classes 3 and 5 in CIFAR10, MNIST, Fashion-MNIST and SVHN datasets. Figure 6 shows the lower bounds for various values of the variance σ used for the Gaussian mixture, where σ is half of the mean distance between data points from the two distributions. |
| Researcher Affiliation | Academia | 1Department of Electrical and Computer Engineering, Univer sity of Wisconsin-Madison, Madison, WI, USA. Correspondence to: Muni Sreenivas Pydi <pydi@wisc.edu>. |
| Pseudocode | No | The paper describes algorithmic strategies but does not include any structured pseudocode or algorithm blocks (e.g., 'Algorithm 1'). |
| Open Source Code | Yes | The code accompanying Figure 6 is made avail able at http://github.com/munisreenivas/ adv-risk-optimal-transport. |
| Open Datasets | Yes | Using this methodology, we evaluate the optimal risk for 2 and adversaries for classes 3 and 5 in CIFAR10, MNIST, Fashion-MNIST and SVHN datasets. |
| Dataset Splits | No | The paper mentions using subsets of datasets (e.g., 'randomly sampling 5000 data points from each class') for analysis of optimal risk bounds, but it does not specify explicit training, validation, and test splits (e.g., percentages or counts) for model training or evaluation needed for reproducibility. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper does not specify any software dependencies with version numbers (e.g., library names with specific versions). |
| Experiment Setup | No | The paper describes parameters for its analysis (e.g., 'randomly sampling 5000 data points from each class', 'variance σ for Gaussian mixture models') but does not provide specific experimental setup details such as hyperparameters, optimizers, or training settings for a machine learning model. |