A sequential linearization method designs step sizes whose Performance Estimation bounds are competitive with BNB-PEP and give accelerated fitted rates for coordinate, inexact, and cyclic gradient descent.
Katyusha: The first direct acceleration of stochastic gradient methods
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
math.OC 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Numerical Design of Optimized First-Order Algorithms
A sequential linearization method designs step sizes whose Performance Estimation bounds are competitive with BNB-PEP and give accelerated fitted rates for coordinate, inexact, and cyclic gradient descent.