Steven Loscalzo
Air Force Research Laboratory
Network
Latest external collaboration on country level. Dive into details by clicking on the dots.
Publication
Featured researches published by Steven Loscalzo.
knowledge discovery and data mining | 2008
Lei Yu; Chris H. Q. Ding; Steven Loscalzo
Many feature selection algorithms have been proposed in the past focusing on improving classification accuracy. In this work, we point out the importance of stable feature selection for knowledge discovery from high-dimensional data, and identify two causes of instability of feature selection algorithms: selection of a minimum subset without redundant features and small sample size. We propose a general framework for stable feature selection which emphasizes both good generalization and stability of feature selection results. The framework identifies dense feature groups based on kernel density estimation and treats features in each dense group as a coherent entity for feature selection. An efficient algorithm DRAGS (Dense Relevant Attribute Group Selector) is developed under this framework. We also introduce a general measure for assessing the stability of feature selection algorithms. Our empirical study based on microarray data verifies that dense feature groups remain stable under random sample hold out, and the DRAGS algorithm is effective in identifying a set of feature groups which exhibit both high classification accuracy and stability.
knowledge discovery and data mining | 2009
Steven Loscalzo; Lei Yu; Chris H. Q. Ding
Stability is an important yet under-addressed issue in feature selection from high-dimensional and small sample data. In this paper, we show that stability of feature selection has a strong dependency on sample size. We propose a novel framework for stable feature selection which first identifies consensus feature groups from subsampling of training samples, and then performs feature selection by treating each consensus feature group as a single entity. Experiments on both synthetic and real-world data sets show that an algorithm developed under this framework is effective at alleviating the problem of small sample size and leads to more stable feature selection results and comparable or better generalization performance than state-of-the-art feature selection algorithms. Synthetic data sets and algorithm source code are available at http://www.cs.binghamton.edu/~lyu/KDD09/.
distributed autonomous robotic systems | 2018
Daniel S. Brown; Ryan Turner; Oliver Hennigh; Steven Loscalzo
Emergent collective behaviors have long interested researchers . These behaviors often result from complex interactions between many individuals following simple rules. However, knowing what collective behaviors are possible given a limited set of capabilities is difficult. Many emergent behaviors are counter-intuitive and unexpected even if the rules each agent follows are carefully constructed. While much work in swarm robotics has studied the problem of designing sets of rules and capabilities that result in a specific collective behavior, little work has examined the problem of exploring and describing the entire set of collective behaviors that can result from a limited set of capabilities. We take what we believe is the first approach to address this problem by presenting a general framework for discovering collective emergent behaviors that result from a specific capability model. Our approach uses novelty search to explore the space of possible behaviors in an objective-agnostic manner. Given this set of explored behaviors we use dimensionality reduction and clustering techniques to discover a finite set of behaviors that form a taxonomy over the behavior space. We apply our methodology to a single, binary-sensor capability model. Using our approach we are able to re-discover cyclic pursuit and aggregation, as well as discover several behaviors previously unknown to be possible with only a single binary sensor: wall following, dispersal, and a milling behavior often displayed by ants and fish.
international conference industrial, engineering & other applications applied intelligent systems | 2015
Jeffrey Hudack; Nathaniel Gemelli; Daniel S. Brown; Steven Loscalzo; Jae C. Oh
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.
algorithmic decision theory | 2015
Daniel S. Brown; Steven Loscalzo; Nathaniel Gemelli
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.
european conference on machine learning | 2013
Robert Wright; Steven Loscalzo; Philip Dexter; Lei Yu
Approximate value iteration methods for reinforcement learning (RL) generalize experience from limited samples across large state-action spaces. The function approximators used in such methods typically introduce errors in value estimation which can harm the quality of the learned value functions. We present a new batch-mode, off-policy, approximate value iteration algorithm called Trajectory Fitted Q-Iteration (TFQI). This approach uses the sequential relationship between samples within a trajectory, a set of samples gathered sequentially from the problem domain, to lessen the adverse influence of approximation errors while deriving long-term value. We provide a detailed description of the TFQI approach and an empirical study that analyzes the impact of our method on two well-known RL benchmarks. Our experiments demonstrate this approach has significant benefits including: better learned policy performance, improved convergence, and some decreased sensitivity to the choice of function approximation.
Archive | 2008
Steven Loscalzo; Lei Yu
Social network analysis can provide great insights into systems composed of interacting objects, and have been successfully applied to various domains. With many different ways to analyze social networks, no single tool currently supports all analysis tasks, but some incorporate more functionality than others. Moreover, the emergence of a new class of social network analysis techniques, link mining, presents a new range of analysis support to provide by the tools. This paper introduces representative social network analysis tasks from traditional, link mining, and visualization aspects, and evaluates a set of tools with diverse general characteristics and social network analysis functionality.
international conference on data mining | 2012
Steven Loscalzo; Robert Wright; Kevin Acunto; Lei Yu
Autonomous agents are emerging in diverse areas and many rely on reinforcement learning (RL) to learn optimal control policies by acting in the environment. This form of learning generates large amounts of transition dynamics data, which can be mined to improve the agents understanding of the environment. There could be many uses for this data, here we focus on mining it to identify a relevant feature subspace. This is vital since RL performs poorly in high-dimensional spaces, such as those that autonomous agents would commonly face in real-world problems. This paper demonstrates the necessity and feasibility of integrating data mining into the learning process while an agent is learning, enabling it to learn to act by both acting and understanding. Doing so requires overcoming challenges regarding data quantity and quality, and difficulty measuring feature relevance with respect to the control policy. We propose the progressive mining framework to address these challenges by relying on cyclic interaction between data mining and RL. We show that a feature selection algorithm developed under this framework, PROFESS, can improve RL scalability better than a competing approach.
genetic and evolutionary computation conference | 2012
Steven Loscalzo; Robert Wright; Kevin Acunto; Lei Yu
international conference on agents and artificial intelligence | 2016
Robert Wright; Steven Loscalzo; Lei Yu