Uniform Convergence, Adversarial Spheres and a Simple Remedy
Authors: Gregor Bachmann, Seyed-Mohsen Moosavi-Dezfooli, Thomas Hofmann
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We validate our theoretical results using numerical experiments on the adversarial spheres dataset. Moreover, we explore the hypothesis put forth by Nagarajan & Kolter (2019b) suggesting that there may exist a decomposition of the model into a clean and a noisy part. We verify our predictions numerically by plotting γK for different kernels and comparing them with the averaged case in Figure 4. |
| Researcher Affiliation | Academia | 1Department of Computer Science, ETH Z urich. Correspondence to: Gregor Bachmann <gregor.bachmann@inf.ethz.ch>. |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any concrete access to source code for the methodology described, such as a repository link or an explicit code release statement. |
| Open Datasets | No | The paper describes a synthetic dataset ('adversarial spheres') and its construction ('Consider the following simple dataset described by the input data distribution p(x) = qpr1(x) + (1 q)pr2(x)'), but it does not provide concrete access information (link, DOI, formal citation) for a publicly available or open dataset. |
| Dataset Splits | No | The paper mentions 'train, test and adversarial accuracy' and discusses 'sample size n' but does not provide specific dataset split information (percentages, sample counts, or citations to predefined splits) needed to reproduce the data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details (exact GPU/CPU models, processor types, or memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment. |
| Experiment Setup | Yes | We use a two 100-dimensional spheres with radii r1 = 1 and r2 = 1.11, similar to the setup considered in Nagarajan & Kolter (2019b). In Figure 6 we show the accuracies of a 2 hidden layer network of width 1000 plotted against different bias initialization magnitudes. |