1. Automated Changes of Problem Representation Eugene Fink LTI Retreat 2007

2. Outline Research interests and projects Concept of representation Representation-changing system Future research challenges Research interests and projects

3. 1999–2003 Assistant Professor 1992–1998 Ph.D. Student Research interests and projects 2004–2007 Systems Scientist

4. Research interests Artificial intelligence Machine learning Algorithm theory Computational geometry Minor Major

5. Main projects Artificial intelligence Machine learning Algorithm theory Computational geometry Representation changes in AI search (Prodigy) Exchange markets for complex commodities Optimization and elicitation under uncertainty (Radar) Generalized convexity Indexing of massive approx- imate data (Argus/Rapid) Medical expert systems Indexing of time series

6. Research themes Representation changes in AI search (Prodigy) Exchange markets for complex commodities Optimization and elicitation under uncertainty (Radar) Generalized convexity Indexing of massive approx- imate data (Argus/Rapid) Medical expert systems Indexing of time series Automated improve- ment of representations Heuristic search in indexing trees Collaboration with the human user Efficient reasoning under uncertainty

7. Research themes Automated improve- ment of representations Representation changes in AI search (Prodigy) Exchange markets for complex commodities Optimization and elicitation under uncertainty (Radar) Generalized convexity Indexing of massive approx- imate data (Argus/Rapid) Medical expert systems Indexing of time series Heuristic search in indexing trees Collaboration with the human user Efficient reasoning under uncertainty Automated improve- ment of representations

8. Outline Research interests and projects Concept of representation Representation-changing system Future research challenges

9. Concept of representation Informally, a representation is a specific approach to solving a problem or class of problems. Psychologists and computer scientists have accumulated evidence on the importance of appropriate representations for both human problem solvers and software systems. Examples: Efficient indexing structures vs. linear search Java vs. assembly language Well written vs. poorly written papers

10. Motivation When humans works on a complex problem, they may need to search for the right approach to the problem. Aha! The related AI challenge is to automated the search for the right approach.

11. Definition of representation A representation consists of three parts: Problem or class of problems Data structures for representing them Algorithms operating on these structures A representation-changing system should select or construct appropriate data structures and algorithms for each given class of problems. problems data structures algorithms

12. Outline Research interests and projects Concept of representation Representation-changing system Future research challenges

13. Representation-changing system Automated representation improvements in the Prodigy problem-solving architecture. Three main parts: Library of search modules Library of learning modules Top-level control mechanism

14. Architecture

15. Manual control Automatic control Control center Given a new problem: Select appropriate modules Apply them to solve the problem Repeat if necessary

16. Top-level control Which learner to apply? Which past results to use? Solve or skip the problem? With which search module? Which learned data to use? Invoke the selected learning module Invoke the selected search module Wait for the next problem Solve the problem or learn additional knowledge?

17. Performance example order of solving problems Solving a series of 50 problems, and improving the related representation.

18. Performance example order of solving problems Solving a series of 50 problems, and improving the related representation.

19. Performance example order of solving problems Solving a series of 500 problems, and improving the related representation.

20. Related publications Eugene Fink. Changes of problem representation. Springer-Verlag, 2003. Eugene Fink. Automatic evaluation and selection of problem solving methods. JETAI journal, 16(2), 2004. Eugene Fink. Evaluation of representations. IEEE SMC conference, under review.

21. Applications Exchange markets for complex commodities: Automated selection of indexing structures, depending on the type and number of orders Elicitation under uncertainty (Radar): Automated selection among elicitation algorithms and related heuristics

22. Outline Research interests and projects Concept of representation Representation-changing system Future research challenges

23. Evaluation of representations Automated selections among data structures and algorithms Automated construction of new data structures Integrated AI systems Applications

24. Evaluation of representations Standard methods for the evaluation and comparison of alternative representations Theory of representation efficiency, which should account for the search time, solution quality, and percentage of solved problems Inherent complexity of AI problems, complexity classes, and AI-completeness

25. Automated selection among data structures and algorithms Combining exploration with exploitation Analysis of similarities among problems and among representations Generation and use of small test problems

26. Automated construction of new data structures Automated selection among alternative data structures Self-adjusting data structures Automated construction of complex structures from basic blocks

27. Integrated AI systems General architecture for integrating search and learning algorithms Large library of standard AI algorithms and tools for their synergetic use The grand challenge is to develop an architecture that integrates thousands of AI engines and problem domains, in the same way as an operating system integrates file-processing programs.

28. Applications The automated improvement of representations should be applicable to most areas of computer science.