Browsing by Author "Dr. Munindar Singh, Committee Member"

Now showing 1 - 10 of 10

Cache Invalidation and Propagation of Updates in Distributed Caching.
(2004-12-16) Kohli, Pooja; Dr. Ioannis (Yannis) Viniotis, Committee Member; Dr. Rada Y. Chirkova, Committee Chair; Dr. Munindar Singh, Committee Member
Replication and caching strategies are increasingly being used to improve performance and reduce user perceived delays in distributed environments. A query can be answered much faster by accessing a cached copy than by making a database roundtrip. This setting creates a number of important issues such as maintaining consistency among copies of the same data item. Numerous techniques have been proposed to achieve caching and replication while maintaining consistency among the replicas. A closer investigation of these schemes reveals that no one scheme can be optimal for all environments. In this thesis we look at invalidation protocols for achieving consistency in systems that use distributed caching. We propose heuristics for dynamic adaptation of these protocols for cache consistency. These heuristics aim at propagating invalidations while reducing the cost of data transfer.
Evaluation of Selected Sensitivity Analysis Methods Based upon Applications to a Probabilistic Food Safety Process Risk Model: Case Study for Listeria monocytogenes
(2003-12-01) Danish, Tanwir; Dr. Christopher Frey, Committee Chair; Dr. Ranji Ranjithan, Committee Member; Dr. Munindar Singh, Committee Member
With the emergence of large food safety risk models, there has also been a growing recognition of the need for sensitivity analysis of such models. Key questions that must be addressed in performing sensitivity analysis with food safety risk models include the following: What are the key criteria for sensitivity analysis methods applied to food safety risk assessment?; What sensitivity analysis methods are most promising for application to food safety and risk assessment?; and What are the key needs for implementation and demonstration of such methods? Food safety risk assessment models are challenging because they typically include: (1) nonlinearities; (2) thresholds; (3) continuous, discrete, and categorical inputs; and (4) two-dimensional simulation of variability and uncertainty. In June 2001, NCSU hosted a workshop on sensitivity analysis. Recommendations were made regarding a guideline document to assist practitioners in selecting, applying, interpreting, and reporting the results of sensitivity analysis. The workshop also supported the need for case studies with existing food safety risk models to demonstrate to practitioners how sensitivity analysis methods can be used and to evaluate various specific methods. The two main purposes of this report are to: (1) evaluate sensitivity analysis methods; and (2) present an example of how sensitivity analysis can be applied to food safety risk assessment models and how the results can be presented and interpreted. This study included a review of existing methods, and a detailed series of quantitative case studies of multiple sensitivity analysis methods applied to Listeria monocytogenes model. Methods evaluated include local sensitivity analysis (e.g., nominal range sensitivity, differential sensitivity), global sensitivity analysis methods (e.g., linear regression, analysis of variance, and regression trees), and the graphical method of scatter plot. The report present example results and insights from the application of these methods of sensitivity analysis to the model. The methods were compared and evaluated with regard to multiple criteria, including applicability, robustness of results, ability to rank and discriminate among inputs, and others. The guidance emerging from this research includes: (1) why and when to perform sensitivity analysis; (2) preparation of existing or new models to facilitate sensitivity analysis; (3) defining objectives and case study scenarios; (4) selection of methods; (5) general principles for application of methods; and (6) presentation and interpretation of results. [This work was supported by a cooperative agreement with U.S. Department of Agriculture's Office of Risk Assessment and Cost-Benefit Analysis]
Identifying Bidding Strategies on eBay: A Feasibility Study
(2002-05-24) Shah, Harshit; Dr. Munindar Singh, Committee Member; Dr. Peter Wurman, Committee Chair; Dr. Jon Doyle, Committee Member
Millions of people participate in online auctions on websites such as eBay. The data available in these public markets offer interesting opportunities to study internet auctions. The main purpose of this research is to identify common bidding patterns that appear on eBay. We examine data from eBay videogame console auctions. A new way of interpreting bidding behaviors is proposed. The analysis reveals that there are certain bidding behaviors that appear frequently in the data. We identify the behaviors and infer bidder's strategy that might lead to such behaviors.
Image Processing for Cognitive Models in Dynamic Gaming Environments
(2003-07-07) Shah, Kunal Deepak; Dr. Robert St. Amant, Committee Chair; Dr. Michael Young, Committee Member; Dr. Munindar Singh, Committee Member
Cognitive models have typically dealt with environments that are either artificial or real but too simplistic. This stems from the fact that the process of describing the environment to the cognitive model is a complex vision problem. In order to realize the full potential of cognitive models, it is imperative that they be able to operate in natural domains. We attempt to overcome this limitation by providing a perceptual component to a cognitive model that interacts with more realistic environments. This perceptual component is an image processing substrate that has been customized for two different gaming environments. The substrate formerly worked only for the static environments we associate with conventional graphical user interfaces; the work we describe here extends its functionality to a more general class of interfaces, as represented by the driving game and Mars rover game. A cognitive model built on the ACT-R cognitive architecture has been developed that demonstrates the use of the image processing substrate in performing the driving task.
On the relative advantages of teaching Web services in .NET vs. J2EE.
(2003-09-17) Kachru, Sandeep; Dr. Munindar Singh, Committee Member; Dr. Edward Gehringer, Committee Chair; Randy Miller, Committee Member; Dr. Laurie Williams, Committee Member
.NET and J2EE are currently the two leading technologies in enterprise-level application development. In the coming years, according to various surveys, these two technologies will capture an almost equal amount of market share. They are also the platforms of choice for developing Web services. There is an ongoing debate about the advantage of developing Web services in one over the other. We look at this question from the perspective of educators. We compare and analyze the two platforms using a number of parameters such as features present in each platform, tools and resources offered by the two and compatibility with the rest of the curriculum. We study the most significant difference between the two platforms — the platform independence of J2EE and the language independence of .NET, and discuss their relative advantages in an academic environment. We discover that both of the platforms offer equal support for the development of Web services and teach the concepts equally well. While .NET offers integrated, native support for various phases of Web services development, Java platform achieves this with several new libraries. On the other hand, J2EE's major advantage over .NET is the popularity of the Java language in academia. Thus, teaching Web services in Java maintains uniformity in the curriculum. A looming factor is the growth of C# as a teaching language. Though it seems destined to be adapted as a primary language in more schools, it will be some time before it can challenge Java as the most popular language in universities. We finally compare the development process of Web services in IBM's Websphere and Microsoft?s Visual Studio .NET and find them remarkably similar. Both the tools provide comparable features to develop Web services easily. Thus, the choice of platform will depend on factors other than the relative ease of teaching Web services. Arguments in favor of J2EE are platform independence, multiple vendor support, popularity of Java in universities, a greater number of tools and resources etc. However, it does not allow programming in any other language besides Java and does not offer native support for Web services. On the other hand, the .NET platform has support for multiple languages, integrated support for Web services, an excellent development tool and a language that is becoming more popular in academia. The factors that go against .NET are inadequate platform independence and single-vendor support. We conclude that there is no clear winner and the choice of platform will depend on various local factors. Finally, we provide a road-map that will help the educators in making the decision.
A Simple Auction for Early Delivery of High-Demand Products
(2004-11-01) Korrapati, Sameer; Dr. Michael Young, Committee Member; Dr. Munindar Singh, Committee Member; Dr. Peter Wurman, Committee Chair
Companies face an important challenge in deciding the allocation and pricing of high-demand products when they can supply them only in limited quantities, especially during initial releases. Instead of traditional means, letting free market mechanisms like auctions solve this problem can lead to more efficient outcomes. The Generalized Vickrey Auction (GVA) occupies a prominent place in auction theory because of its allocative efficiency and incentive properties. But implementing the GVA requires the auctioneer to solve n + 1 optimization problems, where n is the number of bidders. Moreover, it also requires complete information revelation by bidders, which can be a severe constraint in a real world application. Iterative auctions have been identified as a good alternative for practical implementations, though sometimes trading practicality with efficiency. We present a simple iterative auction that lets companies efficiently price and allocate limited quantities of high-demand goods. Self-interested and myopic agents can participate in a distributed manner in this online auction. We consider a supply of multiple units of heterogeneous goods under unit demand by agents that have private valuations. Because of its iterative nature, our auction computes the efficient outcome with very little information revelation from the agents. We show that this outcome is efficient and has incentive compatibility properties, like the GVA, under certain assumptions about the agent valuations. Our simulations have shown that the mechanism converges quickly to the equilibrium and collects relatively little information from the agents.
Strategic Deception in Agents
(2004-08-02) Christian, David Benjamin; Dr. James Lester, Committee Member; Dr. R. Michael Young, Committee Chair; Dr. Munindar Singh, Committee Member
Despite its negative ethical connotations, deception is a useful tool for human social interaction, and plays an important role in the process of creating the stories that pervade our popular culture. In this paper, we describe the deception planner, an implementation of a model of strategic deception. Strategic deception is deception performed in order to achieve or enable some higher goal, as opposed to deception that is performed for the sake of deceiving, or for an unstated purpose. Given a model of a deceiver holding ulterior goals, a model of the goals and abilities of a target agent to be deceived, and a model of the relevant pieces of the world, the deception planner generates a set of statements about the current world state which may be either true or false. Those statements are communicated to the target agent, which updates its world state to reflect this new information. The target then performs planning to achieve its own goals, with no knowledge of the deceiver's goals. If the deception planner generates an successful set of statements, the target agent will create a plan that achieves the deceiver's ulterior goals despite no knowledge of those goals. To find the set of statements that will generate this desired behavior from the target agent, the deception planner models the target agent's planning process. The deception planner searches for a plan that achieves the target agent's goals as well as the deceiver's ulterior goals. When such a plan is found, it is labeled the candidate plan, and the deceiver gives the target agent enough (dis)information so that, given the target agent's knowledge of the world, she can generate that plan. The candidate plan may depend on lies, chosen by the deception planner. The planner ensures that any lies told are not discovered before the target executes enough of the plan to achieve the ulterior goals. Once a candidate plan has been found, the deception planner the finds and counters competing plans. Competing plans are plan that achieve the target's goals but not the deceiver's goals, but are of equivalent or better quality than the candidate plan according to some metric shared by the target and deceiving agent. Because they are of the same or better quality, the target agent may choose a competing plan instead of the candidate plan. A competing plan is countered by undermining through lying a belief that is necessary for that plan to be executable. Although the deception planner fits within a body of work on agent deception, the goal of this algorithm is unique in its focus on causing a target agent to act in order to achieve a deceiver's goals, and in its utilization of a model of the goals and planning abilities of the target agent to that end.
Techniques For Finding Nash Equilibria In Combinatorial Auctions
(2005-07-07) Sureka, Ashish; Dr. Peter Wurman, Committee Chair; Dr. Munindar Singh, Committee Member; Dr. Michael Young, Committee Member; Dr. David Thuente, Committee Member
Auctions that allow participants to bid on a combination of items rather than just the individual items are called combinatorial auctions. For items that exhibit complementarity and substitutability, combinatorial auctions can be used to reach economically efficient allocations of goods and services. There has been a surge of recent research on combinatorial auctions because of the wide variety of practical situations to which they can be applied. There are several instances in which combinatorial auctions have already been applied to allocate scares resources, but there are still some challenging issues that need to be addressed before combinatorial auctions can be much more widely used in practice. Many different combinatorial auctions designs have been proposed by researchers and recently there has been a lot of work on studying the computational and strategic aspects of these auction designs. In this thesis, I analyze combinatorial auctions from a game theoretic perspective and propose techniques for determining pure strategy Nash equilibrium of combinatorial auctions. For a variety of reasons, combinatorial auctions pose serious computational challenges to compute Nash equilibria using current techniques. One problem is that the size of the strategy space in combinatorial auctions is very large and grows exponentially with the number of bidders and items. Another computational issue is that for combinatorial auctions it is computationally expensive to compute the payoffs of the players as a result of the joint actions. This makes it computationally expensive to determine the complete payoff matrix upfront and then determine Nash equilibrium. In this dissertation, we present techniques to overcome these problems. We present algorithms based on meta-heuristic search techniques, best response dynamics and linear programming to tackle these problems. We present empirical and theoretical results to support our claim that the algorithms perform well.
User Interface Softbots
(2003-12-11) Dudani, Ajay; Dr. Munindar Singh, Committee Member; Dr. Michael Young, Committee Member; Dr. Robert St. Amant, Committee Chair
A user interface softbot is a software agent that controls an interactive system through its graphical user interface, relying on visual information of the system rather than an application programming interfaces or access to source code. We have developed a prototype system that facilitates the use and extension of interface softbots, reforming the programming process and making it easier for new developers. We present the JSegMan substrate that facilitates the representation and identification of the conventional graphical user interface. The JSegMan substrate consists of sensors, effectors and framework for an agent that can control the user interface environment. Using image processing the sensor builds a representation of the desktop's visual interface. The effector module generates mouse and keyboard gestures to control the desktop environment. The sensor and effector modules work as eyes and hands of a controller application that can be tailored for a given application domain. We also present three applications - MSN Chatbots, the classical Blocks World problem solver and a GUI testing tool that use the JSegMan substrate and demonstrates applicability of JSegMan in use with off-the shelf applications, PDDL based planners and in testing environment respectively.
Visualization Search Strategies
(2004-11-22) Mehta, Reshma Girish; Dr. Munindar Singh, Committee Member; Dr. Peng Ning, Committee Member; Dr. Christopher Healey, Committee Chair
Innovations in high performance computing and high bandwidth networks have led to the onset of data explosion. Along with large size, the datasets are typically multivariate. The need for effective exploration of this data has led to the area of multidimensional visualization. Research in low level human visual system has resulted in the construction of perceptual guidelines that can produce effective visualizations. However, application of these guidelines to a dataset requires users to be experts in the visualization domain. ViA is a semi-automated visualization assistant that uses perceptual guidelines along with a heuristic search algorithm to generate perceptually salient visualizations. This thesis aims to study the behavior of the current hint-based search strategy and determine its efficiency. We compare hint-based search with two generic heuristic search algorithms, simulated annealing and reactive tabu search, by adapting them to ViA's search domain. We use time efficiency, space efficiency, ability to find multiple optimal solutions and optimality as performance metrics. Further, in order to "see" the areas of the search space explored by each search algorithm, we have developed a focus + context visualization system using hyperbolic geometry.