Improved techniques for deterministic l2 robustness

Authors: Sahil Singla, Soheil Feizi

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

Reproducibility Variable Result LLM Response
Research Type Experimental Using these methods, we significantly advance the state-of-the-art for standard and provable robust accuracies on CIFAR-10 (gains of +1.79% and +3.82%) and similarly on CIFAR-100 (+3.78% and +4.75%) across all networks. Code is available at https://github.com/singlasahil14/improved_l2_robustness.
Researcher Affiliation Academia Sahil Singla Department of Computer Science University of Maryland ssingla@umd.edu Soheil Feizi Department of Computer Science University of Maryland sfeizi@umd.edu
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/singlasahil14/improved_l2_robustness.
Open Datasets Yes We perform experiments under the setting of provably robust image classification on CIFAR-10 and CIFAR-100 datasets.
Dataset Splits Yes Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes]
Hardware Specification Yes All experiments were performed using 1 NVIDIA GeForce RTX 2080 Ti GPU.
Software Dependencies No The paper does not provide specific ancillary software details with version numbers.
Experiment Setup Yes All networks were trained for 200 epochs with initial learning rate of 0.1, dropped by a factor of 0.1 after 100 and 150 epochs. For adversarial training with curvature regularization, we use ρ = 36/255 (0.1411), γ = 0.5 for CIFAR-10 and ρ = 0.2, γ = 0.75 for CIFAR-100.