Online Gradient Ascent
An online optimization algorithm used in data-stream settings to incrementally update parameters so as to maximize an objective function.
Timeline (last 90 days)
In the i.i.d. data stream setting, an online gradient ascent algorithm projected on fixed nodes can produce certificates with a fixed horizon, whose error is bounded by three parts—the grid term, the learning term, and the statistical term—and a predictable betting strategy can give certificates valid at arbitrary times.
Classical results show that under convex losses, the cumulative regret of projected online gradient ascent is bounded by O(√T), where T is the number of rounds (Zinkevich, 2003).
All Facts (2)
In the i.i.d. data stream setting, an online gradient ascent algorithm projected on fixed nodes can produce certificates with a fixed horizon, whose error is bounded by three parts—the grid term, the learning term, and the statistical term—and a predictable betting strategy can give certificates valid at arbitrary times.
50%UnverifiedClassical results show that under convex losses, the cumulative regret of projected online gradient ascent is bounded by O(√T), where T is the number of rounds (Zinkevich, 2003).
50%