Reproducibility Report: Rethinking Softmax Cross-Entropy Loss for Adversarial Robustness

By Reproducibility Summary. 2021

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Boosting Adversarial Training with Hypersphere Embedding (2020, methodology): “WN normalizes the weights in the softmax layer to encourage the model to equally treat samples from different classes, facilitating the cases with unbalanced training data (Guo & Zhang, 2017; Liu et al., 2017; Pang et al., 2018; 2020).”
Rethinking Semantic Segmentation: A Prototype View (2022, background): “Third, with the cross-entropy loss, only the relative relations between intra-class and inter-class distances are optimized [89, 111, 134]; the actual distances between pixels and prototypes, i.”
On Adaptive Attacks to Adversarial Example Defenses (2020, methodology): “…2019) Section 7 Generative Classiﬁers (Li et al., 2019) Section 8 Sparse Fourier Transform (Bafna et al., 2018) Section 9 Rethinking Cross Entropy (Pang et al., 2020a) Section 10 Error Correcting Codes (Verma & Swami, 2019) Section 11 Ensemble Diversity (Pang et al., 2019) Section 12 EMPIR (Sen…”
Advances in adversarial attacks and defenses in computer vision: A survey (2021, methodology): “[162] suggested to replace the softmax cross-entropy loss with a new loss, called MaxMahalanobis center loss to induce adversarial robustness in the model.”
Exploring Cross-Image Pixel Contrast for Semantic Segmentation (2021, methodology): “2) Due to the use of softmax, the loss only depends on the relative relation among logits and cannot directly supervise on the learned representations [56].”
Reliable evaluation of adversarial robustness with an ensemble of diverse parameter-free attacks (2020, unknown): “66 11 (Pang et al., 2020) 80.”
Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack (2019, methodology): “We consider the defense proposed in (Pang et al., 2020), in particular their ResNet-110 (without adversarial training) for CIFAR-10.”
Better Diffusion Models Further Improve Adversarial Training (2023, methodology): “…data, several enhancements for AT have been made employing strategies inspired by other areas, including metric learning (Mao et al., 2019; Pang et al., 2020a;b), self-supervised learning (Chen et al., 2020a;b; Naseer et al., 2020; Wang & Liu, 2022), ensemble learning (Tramèr et al.,…”
Deep Repulsive Prototypes for Adversarial Robustness (2021, methodology): “Pang et al. (2019) introduced a loss function to enforces inter-class separability using the centroids of the Max-Mahalanobis distribution.”