On Explaining Neural Network Robustness with Activation Path

Authors: Ziping Jiang

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 EXPERIMENTS
Researcher Affiliation Academia Ziping Jiang School of Computing and Communications, Lancaster University {z.jiang7}@lancaster.ac.uk
Pseudocode Yes Algorithm 1 Smoothed Classifier with Repressed Float Path
Open Source Code Yes Code is provided at: https://github.com/Orange Bai/APCT-master
Open Datasets Yes For the CIFAR10 dataset, we compare our method with the benchmark classifier proposed by Cohen et al. (2019) with VGG16 network to show the effectiveness of our method. For the Image Net dataset, we choose Res Net50 as model architecture...
Dataset Splits No For the CIFAR10 dataset, we compare our method with the benchmark classifier proposed by Cohen et al. (2019) with VGG16 network to show the effectiveness of our method. For the Image Net dataset, we choose Res Net50 as model architecture and add the adversarial smoothed classifier Salman et al. (2019) as the base model. We also compare our model with recent works (Jeong & Shin (2020); Jeong et al. (2021)) to obtain a general evaluation. In line with previous works, we use certifiable accuracy at different radius computed by Cohen et al. (2019) as the metric.
Hardware Specification No The paper does not mention any specific hardware (e.g., GPU model, CPU type, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as 'PyTorch 1.9' or 'Python 3.8'.
Experiment Setup Yes Each of the models is trained for 200 epochs with SGD optimizer and an initial learning rate of 0.1, which decays after 60, 120, and 160 epochs with a rate of 0.2.