Tri did his master thesis under the supervision of Prof. Thomas Eiter in Knowledge Based Systems Group, Institute of Information Systems, Faculty of Informatics,Vienna University of Technology. He also has two co-supervisors: Thomas Krennwallner and Minh Dao-Tran.
The Title of his master thesis: "Top-Down Evaluation Techniques for Modular Nonmonotonic Logic Programs"
You can find here: his presentation slides, his poster, and finally, his master thesis.
What is it all about? Basically it extends the capability of (nonmonotonic) logic programs with modularity aspects. They enable modular paradigm in logic programming.
Read the abstract:
Answer Set Programming (ASP) is a very useful tool for knowledge representation and declarative problem solving. Recently, enabling modularity aspects in ASP has gained increasing interest to help composing (sub-)programs to a combined logic program. Modularity not only allows for problem decomposition, but also facilitates high (code) reusability and provides better support for large-scale projects. Among the contemporary approaches, Modular Nonmonotonic Logic Programs (MLPs) have distinguished strengths, e.g., they allow for mutual recursive calls and utilize predicate symbols as module inputs, resulting in more dynamic problem encodings. MLPs are very expressive and have high computational complexity, thus creating practicable implementations for this formalism is a very challenging task.
In this thesis, we develop TD-MLP, a concrete algorithm for computing answer sets for MLPs. TD-MLP is based on a top-down evaluation technique which considers only relevant module calls. In addition, we have devised an optimization technique that splits module instantiations to avoid redundant recomputation. We have incorporated the optimization technique into the original approach and experiments on a benchmark suite show promising results. Furthermore, we also evaluate the performance of different encodings for different problems, involving modular and ordinary encodings. Experiments show in some cases our modular encoding can outperforms the ordinary ones.
The implementation code can be found here. Using svn, you can check out the code by:
svn co http://dlvhex.svn.sourceforge.net/svnroot/dlvhex/dlvhex/tags/TD-MLP_1.0_thesis-version/