Convergence Rate
A theoretical measure describing the rate at which optimization error decreases with the number of iterations, often expressed in forms such as O(1/T) or O(1/√T).
Timeline (last 90 days)
In adversarial training, there is no uniform convergence rate.
In standard training, gradient descent has a uniform convergence rate guarantee on smooth convex problems, independent of the specific data distribution.
The paper proves that, under appropriate network depth and width configurations, the convergence rate of dense ReLU networks can overcome the curse of dimensionality, achieving polynomial-level sample complexity.
All Facts (3)
In adversarial training, there is no uniform convergence rate.
50%UnverifiedIn standard training, gradient descent has a uniform convergence rate guarantee on smooth convex problems, independent of the specific data distribution.
50%UnverifiedThe paper proves that, under appropriate network depth and width configurations, the convergence rate of dense ReLU networks can overcome the curse of dimensionality, achieving polynomial-level sample complexity.
50%