Critical Computational Infrastructure for Real-World Large Scale Decision-Making Support Systems
Abstract
Abstract A system that understands the decision making process and how each decision is interrelated can improve commanders’ performance by enabling them to make effective and efficient choices ranging from generating the appropriate course of action response to an event, to selecting where to allocate troops or equipment. This combination of user and intent models in a probabilistic system allows modeling the underlying process, while avoiding common pitfalls of Decision Support Systems (DSS) such as integrating poorly within the workflow, being difficult to keep up to date, and being unable to generalize and apply to other problems. A tight integration further facilitates global optimization over objectives from multiple commanders at their own respective levels, identifying important tradeoffs across all goals. This system must be able to react quickly to new information and consider a wide variety of decision options in order to be a crucial part of the commanders’ workflow. In order to adapt quickly, we are proposing to use portfolios of distributed anytime-anywhere algorithms to compute many options simultaneously with a dynamic programming like approach to sharing partial results across computations. This has the potential to significantly scale up computations but requires a large memory platform to facilitate sharing partial results in addition to utilizing multiple computational nodes. Furthermore, a large memory base enables the ability to characterize the overall problem space with better analysis of how the problem structurally grows and ultimately, what hardware is required for any given problem. Realistically, not every computation needs to always reside in memory. Persistent storage for prior results and reports is likely necessary. Thus, we are proposing to procure a system composed of a high memory capacity (8TB) multi-node computational platform and a robust Storage Area Network allowing high throughput access to data while minimizing caching time and ensuring data integrity.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512514
Entities
People
- Eugene Santos
Organizations
- Board of Trustees of Dartmouth College
- Office of Naval Research
- United States Navy