Multi-armed bandit improvement
AssigneeMellanox Technologies, Ltd.
InventorsAmir Rosen, Shie Mannor, Gil Levy, Arye Albahari, Ariel Szapiro
Life of this patent
Estimating the term…
Abstract
In one embodiment, a system includes a memory, and a processor coupled to the memory, where the processor is to explore a plurality of control policies during exploration phases, wherein the control policies are associated with different probabilities of being selected for exploration, measure system performance during the exploration phases, and control a resource during an exploitation phase based on a policy selected to maximize expected reward based on the measured system performance.
Claims20
A system, comprising:
- a memory;
- and
- a processor coupled to the memory, where the processor is to:
- explore a plurality of control policies during exploration phases, wherein the control policies are associated with different probabilities of being selected for exploration;
- measure system performance during the exploration phases;
- and
- control a resource during an exploitation phase based on a policy selected to maximize expected reward based on the measured system performance.
A method of executing a control policy selection mechanism in a processing unit, comprising:
- exploring a plurality of control policies during exploration phases, wherein the control policies are associated with different probabilities of being selected for exploration;
- measuring system performance during the exploration phases;
- and
- controlling a resource during an exploitation phase based on a policy selected to maximize expected reward based on the measured system performance.
Dependencies are inferred from the claim text.