Team learning as a game
A machine FIN-learning machine M receives successive values of the function f it is learning; at some point M outputs conjecture which should be a correct index of f. When n machines simultaneously learn the same function f and at least k of these machines outut correct indices of f, we have team learning [k,n]FIN. Papers [DKV92, DK96] show that sometimes a team or a robabilistic learner can simulate another one, if its probability p (or team success ratio k/n) is close enough. On the other hand, there are critical ratios which mae simulation o FIN(p2) by FIN(p1) imossible whenever p2 _< r < p1 or some critical ratio r. Accordingly to [DKV92] the critical ratio closest to 1/2 rom the let is 24/49; [DK96] rovides other unusual constants. These results are comlicated and rovide a ull icture o only or FIN- learners with success ratio above 12/25.
We generalize [k, n]FIN teams to asymmetric teams [AFS97]. We introduce a two player game on two 0-1 matrices defining two asymmetric teams. The result of the game reflects the comparative power of these asymmetric teams. Hereby we show that the problem to determine whether [k1]FIN ⊂ [k2, n2]FIN is algorithmically solvable. We also show that the set of all critical ratios is well-ordered. Simulating asymmetric teams with probabilistic machines from [AFS97] provides some insight about the unusual constants like 24/49.
Unable to display preview. Download preview PDF.
- [AFS97]K. Apsītis, R. Freivalds, and C. Smith. Asymmetric team learning. Available from http://www.cs.umd.edu/-kalvis/, Accepted to COLT'97.Google Scholar
- [Amb96]A. Ambainis. Probabilistic pfin-type learning: general properties. In Proceedings of the 9th Conference on Computational Learning Theory. ACM, 1996.Google Scholar
- [DK95]R. Daley and B. Kalyanasundaram. Towards reduction arguments for finite learning. Lecture Notes in Artificial 961:63–75, 1995.Google Scholar
- [DK96]R. Daley and B Kalyanasundaram. Finite learning capabilities and their limits. Available from http://www.cs.pitt.edu/~daley/fin/fin.html, 1996.Google Scholar
- [DKV92]R. Daley, B. Kalyanasundaram, and M. Velauthapillai. Breaking the probability 1/2 barrier in fin-type learning. In L. Valiant and M. Warmuth, editors, Proceedings of the Fifth Annual Workshop on Computational Learning Theory, pages 203–217. ACM Press, 1992.Google Scholar
- [DPVW91]R. Daley, L. Pitt, M. Velauthapillai, and T. Will. Relations between probabilistic and team one-shot learners. In M. Warmuth and L. Valiant, editors, Proceedings of the 1991 Workshop on Computational Learning Theory, pages 228–239, Palo Alto, CA., 1991. Morgan Kaufmann Publishers.Google Scholar
- [JS90]S. Jain and A. Sharma.Finite learning by a team.In M. Fulk and J. Case, editors, Proceedings of the Third Annual Workshop on Computational Learning Theory, pages 163–177, Palo Alto, CA., 1990. Morgan Kaufmann Publishers.Google Scholar
- [KS94]M. Kummer and F. Stephan. Inclusion problems in parallel learning and games. In Proc. 7th Annu. ACM Workshop on Comput. Learning Theory, pages 287–298. ACM Press, New York, NY, 1994.Google Scholar
- [NW63]C. Nash-Williams. On well-quasi-ordering finite trees. Proc. Cambridge Phil. Soc., 59:833–853, 1963.Google Scholar
- [PS88]L. Pitt and C. Smith. Probability and plurality for aggregations of learning machines. Information and Computation, 77:77–92, 1988.Google Scholar
- [Smu61]R. Smullyan. Theory of Formal Systems, Annals of Mathematical Studies, volume 47. Princeton University Press, Princeton, New Jersey, 1961.Google Scholar
- [Ve189]M. Velauthapillai. Inductive inference with a bounded number of mind changes. In R. Rivest, D. Haussler, and M. Warmuth, editors, Proceedings of the 1989 Workshop on Computational Learning Theory, pages 200–213, Palo Alto, CA., 1989. Morgan Kaufmann.Google Scholar