Dejvuth Suwimonteerabuth

Symbolic Context-Bounded Analysis of Multithreaded Java Programs

The reachability problem is undecidable for programs with both recursive procedures and multiple threads with shared memory. Approaches to this problem have been the focus of much recent research. One of these is to use context-bounded reachability, i.e. to consider only those runs in which the active thread changes at most ktimes, where kis fixed. However, to the best of our knowledge, context-bounded reachability has not been implemented in any tool so far, primarily because its worst-case runtime is prohibitively high, i.e. O(nk), where nis the size of the shared memory. Moreover, existing algorithms for context-bounded reachability do not admit a meaningful symbolic implementation (e.g., using BDDs) to reduce the run-time in practice. In this paper, we propose an improvement that overcomes this problem. We have implemented our approach in the tool jMoped and report on experiments.

jMoped: a java bytecode checker based on moped

We present a tool for finding errors in Java programs that translates Java bytecodes into symbolic pushdown systems, which are then checked by the Moped tool [1].

Efficient algorithms for alternating pushdown systems with an application to the computation of certificate chains

Motivated by recent applications of pushdown systems to computer security problems, we present an efficient algorithm for the reachability problem of alternating pushdown systems. Although the algorithm is exponential, a careful analysis reveals that the exponent is usually small in typical applications. We show that the algorithm can be used to compute winning regions in pushdown games. In a second contribution, we observe that the algorithm runs in polynomial time for a certain subproblem, and show that the computation of certificate chains with threshold certificates in the SPKI/SDSI authorization framework can be reduced to this subproblem. We present a detailed complexity analysis of the algorithm and its application, and report on experimental results obtained with a prototype implementation.

jMoped: a test environment for java programs

We present jMoped [1], a test environment for Java programs. Given a Java method, jMoped can simulate its execution for all possible arguments within a finite range and generate coverage information for these executions. Moreover, it checks for some common Java errors, i.e. assertion violations, null pointer exceptions, and array bound violations. When an error is found, jMoped finds out the arguments that lead to the error. A JUnit [2] test case can also be automatically generated for further testing.

SDSIrep: a reputation system based on SDSI

We introduce SDSIrep, a reputation system based on the SPKI/SDSI authorization system. It is well-known that a system of SPKI/SDSI certificates corresponds to the formal model of a pushdown system (PDS). Our system, SDSIrep, allows principals to express trust and recommendations in the form of so-called certificates with weights. By interpreting weights as probabilities, we obtain a random-walk model of the reputation of a principal. Thus, SDSIrep represents an application of the theory of probabilistic PDSs to the field of computer security. We present an algorithm to compute the reputation of each principal. An extension of SDSIrep also provides for so-called intersection certificates, by which, loosely speaking, a principal gains reputation if recommended by all members of a given group of principals. On a formal-methods level, this extension makes SDSIrep correspond to probabilistic alternating PDSs, and we extend the underlying theory of PDSs to handle this case. As an example we sketch a small academic reputation system that combines information from different reputation sources, like conferences, coauthors, and rankings.

Online robot learning by reward and punishment for a mobile robot

The existing robot learning methods require specifically defined goals. We aim to produce a more flexible behavior. We present our work which a human observer can influence the robot behavior. The robot learns by reward and punishment from a human in real-time. To examine the developed approach, we perform a control system for a color-following task as an example. A physical robot is used to perform the experiments. Experimental results show the emergence of learned behaviors. We discussed the factors that influence the learning process.