Nathaniel Gemelli

Intelligent systems technology for higher level fusion

The premise of this paper is that a combination of information extraction techniques, knowledge bases and natural language processing technology can assist the intelligence analyst by providing higher level fusion capabilities to support the decision making process. The paper examines programs and the tools that have evolved from these programs being researched by the Air Force Research Laboratorys Information Directorate. These programs include DARPA sponsored High Performance Knowledge Bases (HPKB), Rapid Knowledge Formation (RKF) and Evidence Extraction and Link Discovery (EELD). Some of the tools include the CYC knowledge base, Intelligent Mining Platform for the Analysis of Counter Terrorism (IMPACT) and the START natural language query system. By exploiting and leveraging the strengths of each system, we believe that a high level of information fusion is possible.

A rationality-based modeling for coalition support

We present a game theoretic model for multi-agent resource distribution and allocation where agents in the environment must help each other to survive. Each agent maintains a set of two-tuples T=(A, P) called friendship values representing actual friendship and perceived friendship. The model directly addresses problems in reputation management schemes in multi-agent systems and peer-to-peer distributed systems. We present algorithms for maintaining the friendship values as well as a utility equation used in each agents decision making. For an application problem, we adapted our formal model to the military coalition support problem in peace-keeping missions. Simulation results show that efficient resource allocation and sharing with minimum communication cost is achieved without centralized control.

Capabilities aware routing for dynamic adhoc networks

Our objective is to design and construct a useful routing protocol for an adhoc wireless network. While working primarily in the small, hand-held computing device arena, Bluetooth wireless technology has emerged as the network medium of choice. We introduce Bluetooth wireless technology, examine current routing protocols, and present the objectives and considerations for the design of a new Bluetooth routing protocol. The protocol design will consider the capabilities of the devices (nodes) within the range of the network. It is envisioned that the capabilities aware routing (CAR) protocol will make routing decisions based on such factors as device power constraints, memory, location and signal strength.

Multiobjective Optimization for the Stochastic Physical Search Problem

We model an intelligence collection activity as multiobjective optimization on a binary stochastic physical search problem, providing formal definitions of the problem space and nondominated solution sets. We present the Iterative Domination Solver as an approximate method for generating solution sets that can be used by a human decision maker to meet the goals of a mission. We show that our approximate algorithm performs well across a range of uncertainty parameters, with orders of magnitude less execution time than existing solutions on randomly generated instances.

k-Agent Sufficiency for Multiagent Stochastic Physical Search Problems

In many multi-agent applications, such as patrol, shopping, or mining, a group of agents must use limited resources to successfully accomplish a task possibly available at several distinct sites. We investigate problems where agents must expend resources e.g. battery power to both travel between sites and to accomplish the task at a site, and where agents only have probabilistic knowledge about the availability and cost of accomplishing the task at any location. Previous research on Multiagent Stochastic Physical Search mSPS has only explored the case when sites are located along a path, and has not investigated the minimal number of agents required for an optimal solution. We extend previous work by exploring physical search problems on both paths and in 2-dimensional Euclidean space. Additionally, we allow the number of agents to be part of the optimization. Often, research into multiagent systems ignores the question of how many agents should actually be used to solve a problem. To investigate this question, we introduce the condition of k-agent sufficiency for a multiagent optimization problem, which means that an optimal solution exists that requires only k agents. We show that mSPS along a path with a single starting location is at most 2-agent sufficient, and quite often 1-agent sufficient. Using an optimal branch-and-bound algorithm, we also show that even in Euclidean space, optimal solutions are often only 2- or 3-agent sufficient on average.

Exact and Heuristic Algorithms for Risk‐Aware Stochastic Physical Search

We consider an intelligent agent seeking to obtain an item from one of several physical locations, where the cost to obtain the item at each location is stochastic. We study risk‐aware stochastic physical search (RA‐SPS), where both the cost to travel and the cost to obtain the item are taken from the same budget and where the objective is to maximize the probability of success while minimizing the required budget. This type of problem models many task‐planning scenarios, such as space exploration, shopping, or surveillance. In these types of scenarios, the actual cost of completing an objective at a location may only be revealed when an agent physically arrives at the location, and the agent may need to use a single resource to both search for and acquire the item of interest. We present exact and heuristic algorithms for solving RA‐SPS problems on complete metric graphs. We first formulate the problem as mixed integer linear programming problem. We then develop custom branch and bound algorithms that result in a dramatic reduction in computation time. Using these algorithms, we generate empirical insights into the hardness landscape of the RA‐SPS problem and compare the performance of several heuristics.

Asynchronous chess competition

Asynchronous Chess (AChess) is a platform for the development and evaluation of real-time adversarial agent technologies. It is a two-player game using the basic rules of chess with the modification that agents may move as many pieces as they want at any time. Modifying chess in this way creates a new robust, asynchronous, real-time game in which agents must carefully balance their time between reasoning and acting in order to out-perform their opponent. As a fast-paced adversarial game, many challenges relevant to real-world applications arise which give it merit for study and use.

Virtual Structure Reduction on Distributed K-Coloring Problems

Distributed Constraint Problem solving represents a fundamental research area in distributed artificial intelligence and multi-agent systems. The constraint density, or the ratio of the number of constraints to the number of variables, determines the difficulty of either finding a solution or minimizing the set of variable assignment conflicts in a constraint problem. Reducing the active density of a problem typically reduces difficulty in finding a solution. We present a fully distributed technique for reducing the active density of constraint graphs that represent Distributed Constraint Optimization Problems (DCOP), called Virtual Structure Reduction (VSR). The VSR technique leverages the occurrence of frozen pairs, or variables that must be assigned the same value based on shared constraints between variables within the local neighborhood. We show how frozen pairs can be used to identify surrogate relationships between variables which reduces the active set of nodes in the constraint graph and leads to a reduction in active density. We discuss our DCOP solver, integrated with the Distributed Stochastic Algorithm (DSA), called VSR-DSA. The VSR-DSA algorithm demonstrates significant performance gains compared to the DSA-B and MGM algorithms in messages, cycles, solution quality, and time on randomized instances of the 3-coloring problem.

Adaptive State Space Abstraction Using Neuroevolution

In this paper, we present a new machine learning algorithm, RL-SANE, which uses a combination of neuroevolution (NE) and traditional reinforcement learning (RL) techniques to improve learning performance. RL-SANE is an innovative combination of the neuroevolutionary algorithm NEAT[9] and the RL algorithm Sarsa(λ)[12]. It uses the special ability of NEAT to generate and train customized neural networks that provide a means for reducing the size of the state space through state aggregation. Reducing the size of the state space through aggregation enables Sarsa(λ) to be applied to much more difficult problems than standard tabular based approaches. Previous similar work in this area, such as in Whiteson and Stone [15] and Stanley and Miikkulainen [10], have shown positive results. This paper gives a brief overview of neuroevolutionary methods, introduces the RL-SANE algorithm, presents a comparative analysis of RL-SANE to other neuroevolutionary algorithms, and concludes with a discussion of enhancements that need to be made to RL-SANE.

A vehicle-target simulation model for network-centric joint air operations

Joint air operations (JAO) are traditionally orchestrated using static vehicle roles assigned from command and control. The central command and control model used by the Air Force cannot anticipate changes in the battlespace and take advantage of continuous information provided by the sensors in a network-centric environment. With recent advances in information and communication technology and the increased need for a dynamic and flexible response, vehicles are expected to assume multiple roles over the course of a mission. In this study, we develop a simulation model that considers four competing objectives (effort, effectiveness, efficiency, and connectivity) to assess vehicle-target allocation for network-centric JAO.