Attention Allocation for Decision Making Queues
Abstract
We consider the optimal servicing of a queue with sigmoid server performance. The sigmoid server performance occurs in various domains including human decision making, visual perception, human-machine communication and advertising response. The tasks arrive at a given rate to the server. Each task has a deadline that is incorporated as a latency penalty. We investigate the trade-off between the reward obtained by processing the current task and the penalty incurred due to the tasks waiting in the queue. We study this optimization problem in a Markov decision process (MDP) framework and show that the MDP formulation is equivalent to a certainty-equivalent problem. We determine the receding horizon servicing policy for the queue and show that the optimal policy may drop some tasks, that is, may not process a task at all. We then develop an adaptive policy that incorporates all the available information about the current tasks and show that the adaptive policy improves the performance significantly. Finally, we present a comparative study of the receding horizon policy for the certainty-equivalent problem and the adaptive policy. We also suggest guidelines for the design of such queues.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 22, 2012
- Accession Number
- ADA563580
Entities
People
- Cedric Langbort
- Francesco Bullo
- Ruggero Carli
- Vaibhav Srivastava
Organizations
- University of California, Santa Barbara