Abstract:Using linear methods for reinforcement learning problems requires designing efficient features. However, designing features often requires having ample knowledge about the problem domain. When dealing with complex problem domains, coming up with efficient feature sets often requires a trial and error process which can prove difficult or inefficient. We present an evolutionary algorithm for generating and evaluating candidate feature sets for learning a task using gradient descent Sarsa(λ) as a linear method. Our evaluations on three different problem domains show that our solution is effective.
Reza Mahjourian, "An Architectural Style for Data-Driven Systems," 10th International Conference on Software Reuse (ICSR), 2008
Abstract: Data-driven systems and applications are specialized software solutions for acquisition, management, and presentation of information. These systems are usually developed using the same software tools, technologies, and processes used for creating any other type of software. Not only is this approach inefficient, but also it results in extreme redundancies due to the inherently repetitive nature of these applications. However, data-driven systems exhibit characteristics which can be exploited for extensive reuse across a single application or a family of applications. In this paper, we present XPage, an architectural style which is especially designed for building data-driven systems. We also provide several case studies from real-world deployments of XPage to help evaluate its efficiency and flexibility for developing real-world solutions.
Reza Mahjourian, Feng Chen, Ravi Itwari, My Thai, Hongqiang Zhai, Yuguang Fang, "An Approximation Algorithm for Conflict-Aware Broadcast Scheduling in Wireless Ad Hoc Networks," The ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc), 2008, to appear
Abstract: Broadcast scheduling is a fundamental problem in wireless ad hoc networks. The objective of a broadcast schedule is to deliver a message from a given source to all other nodes in a minimum amount of time. At the same time, in order for the broadcast to proceed as predicted in the schedule, it must not contain parallel transmissions which can be conflicting based on the collision and interference parameters in the wireless network. Most existing work on this problem use a limited network model which accounts only for conflicts occurring inside the transmission ranges of the nodes. The broadcast schedules produced by these algorithms are likely to experience unpredictable delays when deployed in the network. This is because they do not take into consideration other important sources of conflict in parallel transmissions, namely the interference range and the carrier sensing range. In this paper we develop a conflict-aware network model, which uses these parameters to increase the probability of scheduling conflict-free transmissions, and thereby improve the reliability of the broadcast schedule. We present and prove correctness of a constant approximation algorithm for minimum-latency broadcast scheduling under this network model. We also present a greedy heuristic algorithm for the same problem. Experimental results are provided to evaluate the performance of our algorithms. In addition, the algorithms are analyzed to justify their performance trends.
Chris A. Mattmann, David Woollard, Nenad Medvidovic, and Reza Mahjourian, "Software Connector Classification and Selection for Data-Intensive Systems," 2nd International Workshop on Incorporating COTS Software into Software Systems, 2007
Abstract: Data-intensive systems and applications transfer large volumes of data and metadata to highly distributed users separated by geographic distance and organizational boundaries. An influential element in these large volume data transfers is the selection of the appropriate software connector that satisfies user constraints on the required data distribution scenarios. Currently, this task is typically accomplished by consulting "gurus", who rely on their intuitions, at best backed by anecdotal evidence. In this paper we present a systematic approach for selecting software connectors based on eight key dimensions of data distribution that we use to represent the data distribution scenarios. Our approach, dubbed DISCO, has been implemented as a Java-based framework. The early experience with DISCO indicates good accuracy and scalability.