A NEURAL NETWORK PERSPECTIVE ON THE MULTI-SECRETARY PROBLEM
Loading...
Authors
Sigal, Yonatan
Subjects
multi secretary problem
drones
decision policy
neural network
drones
decision policy
neural network
Advisors
Szechtman, Roberto
Date of Issue
2024-09
Date
Publisher
Monterey, CA; Naval Postgraduate School
Language
Abstract
Drones are prevalent in battlespaces around the world due to their easy availability, low cost, and adaptability. In the simplest form, targets can be characterized by their value and availability. Our research aims at finding targeting policies for a drone, carrying limited ammunition and operating over a finite window of time, in order to maximize the expected value of targets destroyed during a mission. The problem is modeled as a multi-secretary problem whereby, given a number of missiles, the goal is to fire at the highest-value targets in a limited amount of time, with the following restrictions: each target is present for one time period and they arrive one at a time independently of one another from the same (known) distribution representative of their value. The novelties of our modeling approach are: (i) we allow the targets values and availability to be Markovian, and (ii) we adapt neural network algorithms to this setting. The neural network framework is flexible enough to incorporate contextual information, such as weather and intelligence, into the drone operation. Time permitting, we will refine the neural network framework so it can handle adversarial settings, where the targets are presented adaptively by a strategic opponent.
Type
Thesis
Description
Series/Report No
Department
Organization
Identifiers
NPS Report Number
Sponsors
Funder
Format
Citation
Distribution Statement
Distribution Statement A. Approved for public release: Distribution is unlimited.
Rights
�Copyright�is reserved by the copyright owner.