Adversarial Linear Logistic Regression
A simplified model for analyzing adversarial training theory: a linear classifier with logistic loss, where the inner adversarial perturbation maximizes the loss within a given norm ball, retaining the core structure of adversarial training.
Timeline (last 90 days)
In adversarial linear logistic regression, the gap in implicit bias between adversarial training and standard training persists even when the step size is arbitrarily large.
Adversarial linear logistic regression is a commonly used simplified model for analyzing adversarial training theory: the model is a linear classifier, the loss is the logistic loss, and the inner adversarial perturbation maximizes the loss within a given norm ball.
All Facts (2)
In adversarial linear logistic regression, the gap in implicit bias between adversarial training and standard training persists even when the step size is arbitrarily large.
50%UnverifiedAdversarial linear logistic regression is a commonly used simplified model for analyzing adversarial training theory: the model is a linear classifier, the loss is the logistic loss, and the inner adversarial perturbation maximizes the loss within a given norm ball.
50%