TECHNOLOGY AREA(S): Info Systems
OBJECTIVE: To develop Machine Learning based scheduling algorithm(s) for sharing resources, which may be geographically distributed, in complex military environments that makes dynamic assignment of resources, to tasks, while accounting for constraints imposed by networked resources.
DESCRIPTION: Sharing platforms in civilian logistics (e.g. Airbnb, Uber/Lyft) have provided novel ways to implement logistics. By improving utilization of untapped resources, these platforms have led to increased efficiency. Recently, the adoption of sharing and pooling platforms has also been considered in military settings (see Braw (2016)). The current fiscal environment calls for more efficient utilization of military resources. However, sharing resources in a military environment exhibits unique features and requirements that pose research and implementation challenges: - Uncertainty: Due to the complexity of military operations, the duration of resource consumption (or utilization) is uncertain and difficult to predict. Civilian platforms such as Airbnb exclusively deal with fixed-duration consumption requests. Uncertainty of resource consumption usually leads to under-utilization of resources and delays in satisfying demand requests. - Networked resources: Instead of a single resource, requests typically pertain to a heterogeneous bundle of resources in a network configuration. In contrast, a car ride or a room rental consumes a single resource. Requests in the form of bundles introduces inter-dependencies between resources as well as other requests. - Risk Management: Sharing platforms for military applications must incorporate risk management considerations that are absent in civilian applications. For example, in mission critical settings a minimum standard for quality of service must be guaranteed. Unlike civilian platforms for resource sharing in which priorities for resource allocation decisions are based on monetary transactions, military sharing platforms must make use of non-monetary mechanisms. The main objective of this project is to develop and analyze implementable, data-driven algorithms that address challenges in managing large-scale sharing platforms with the above characteristics. The focus is on algorithms that match supply and demand, as well as algorithms that assign priorities in a manner consistent with operational doctrine. This type of large-scale algorithms have been elusive so far due to lack of reliable data on past resource consumption patterns. Recent developments in information technologies (e.g., real-time location systems, pervasive communications) and widespread implementation of electronic management systems are generating large volumes of operational data, which the developed approaches should be capable of utilizing.
PHASE I: To conceive, implement and test a new class of reservation algorithms in order to match real time supply and demand in ways that increases the overall system efficiency, as well as provide an opportunity for adequate planning. Viable and accurate schedules resulting from reservation algorithms are essential for efficient operation of resource sharing platforms in accordance with OPTEMPO. To achieve this objective, algorithms and software capable of producing efficient reservation schedules for resource sharing platforms with a large number of resources and requests must be developed (see e.g. Busic and Meyn (2015), Gurvich and Ward (2014)). The proposed set of methods must also incorporate risk and quality-of-service attributes in determining resource allocation. The final deliverable of Phase I will be a simulation testbed illustrating the benefits and trade-offs associated to the proposed algorithms in a specific military setting (e.g. battlefield logistics, sharing UAVs for surveillance and reconnaissance).
PHASE II: In light of significant uncertainty, reservation algorithms based upon historical averages of duration (the current state-of-the-art) will surely lead to significant discrepancies between planned and actual processes. In Phase II, the objective is to develop fast machine learning techniques to complement the algorithms developed in Phase I so as to fully exploit updated information on statistical variability of resource consumption which will be fedback to update scheduling decisions according to algorithms proposed in Phase I. Synthetic data from simulated operations as well as forecasts from planned operations could be used as input for the learning task. An additional objective for Phase II is to develop a networking architecture capable of supporting: -Platform Dynamics: Spatial and temporal movement of resources in the system. In some platforms, the dynamics are primarily temporal (when resources get occupied and freed); in vehicle sharing, they are spatio-temporal. -Resource Constraints: These determine the possible states of the system and the set of allowed matches in each state. In vehicle sharing, the state space is the location and type (free/in-use) of each vehicle; feasible matches require demand to arrive at stations with free vehicles. For other platforms, the state space is whether each resource is free/occupied; allowed matches depend on compatibility between customers and resources. -Request Dynamics: These can be one-sided (typically when supply is fixed), or two-sided (where both demand and supply actively participate).
PHASE III: In Phase III, the software developed in Phases I and II will be made available for military and civilian use (e.g. sharing platform for complex logistic operations). We envision that the team that develops the software will market it for Government laboratory use, and negotiate commercial licensing with commercial and academic markets. As an alternative, any or all of these artifacts might be released into the open source community. Based on negotiations with the types of government and commercial organizations cited, it is possible that hybrid commercial and open source licensing could occur. In the case where these artifacts are released into the open source community, the STTR awardee would need to develop and provide a plan to state how it would sell additional consulting, software implementation and/or training services around their workflow model, technical implementation guidelines, and/or software controls.
1: E. Braw (2016) The Military Sharing Economy. Foreign Affairs, https://www.foreignaffairs.com/articles/germany/2016-03-07/military-sharing-economy.
2: A. Busic and S. Meyn. (2015) Approximate Optimality with Bounded Regret in Dynamic Matching Models. ACM SIGMETRICS Performance Evaluation Review, Vol 43 No. pp.75-90.
3: I. Gurvich and A. Ward (2014) On the Dynamic Control of Matching Queues, Stochastic Systems, Vol.4 pp. 1-4
KEYWORDS: Sharing Platforms, Army Logistics, Real-time Matching