Abstract
We present new procedures for inferring the structure of a finite-state automaton (FSA) from its input/output behavior, using access to the automaton to perform experiments.
Our procedures use a new representation for finite automata, based on the notion of equivalence between tests. We call the number of such equivalence classes the diversity of the automaton; the diversity may be as small as the logarithm of the number of states of the automaton. For the special class of permutation automata, we describe an inference procedure that runs in time polynomial in the diversity and log(1/δ), where δ is a given upper bound on the probability that our procedure returns an incorrect result. (Since our procedure uses randomization to perform experiments, there is a certain controllable chance that it will return an erroneous result.) We also discuss techniques for handling more general automata.
We present evidence for the practical efficiency of our approach. For example, our procedure is able to infer the structure of an automaton based on Rubik's Cube (which has approximately 1019 states) in about 2 minutes on a DEC MicroVax. This automaton is many orders of magnitude larger than possible with previous techniques, which would require time proportional at least to the number of global states. (Note that in this example, only a small fraction (10-14) of the global states were even visited.)
Finally, we present a new procedure for inferring automata of a special type in which the global state is composed of a vector of binary local state variables, all of which are observable (or visible) to the experimenter. Our inference procedure runs provably in time polynomial in the size of this vector (which happens to be the diversity of the automaton), even though the global state space may be exponentially larger. The procedure plans and executes experiments on the unknown automaton; we show that the number of input symbols given to the automaton during this process is (to within a constant factor) the best possible.
- ~ANGLUIN, D. 1978. On the complexity of mlmmum inference of regular sets. Inf. Cont. 39, ~337-350.Google Scholar
- ~ANGLUIN, D. 1982. Inference of reversible languages. J. ACM 29, 3 (July) 741-765. Google Scholar
- ~ANGLUIN, D. 1987a. Learning regular sets from queries and counterexamples. Inf. Computatzon ~75. (Nov), 87-106. Google Scholar
- ~ANGLUIN, m. 1987b. A note on diversity. Unpublished manuscript.Google Scholar
- ~BAINBR1DGE, E. S. 1977. The fundamental duality of system theory. In Systems: .4pproaches, ~Theones, Apphcattons, W. E. Hartnett, Ed. Reidel, Dordrecht, Holland, pp. 45-61Google Scholar
- ~BERMAN, A. AND PLEMMONS, R. J. 1979. Nonnegatwe 3lattices m the Mathematical Sciences. ~Academic Press, Orlando, Fla.Google Scholar
- ~CORMEN, T. H., LEiSERSON, C. E., AND RIVEST, R. L. 1990. hztroductton to/t{gortthms. MIT ~Press, Cambridge, Mass.Google Scholar
- ~DEAN, T., ANGLUIN, D., BASYE, K., ENGELSON, S., KAELBLING, L., KOKKEVIS, E., AND MARON, O. ~1992. Inferring finite automata with stochastic output functions and an application to map ~learning. In Proceedings of the lOth Natzoua/ Confereuce on ,4rtificial bztelhgence (July). MIT ~Press, Cambridge, Mass., pp. 208-214.Google Scholar
- ~DRESCHER, G. L. 1986. Genetic AI--translating Piaget into Lisp. Tech. Rep. 890. MIT Artificial ~Intelligence Laboratory, Cambridge, Mass., Feb. Google Scholar
- ~DRESCHER, G. L. 1987. A mechanism for early Piagetian learning. In Proceedings of AAAI-87.' ~Sixth National Conference on Artificial Intelligence (Seattle, Wash., July). Morgan-Kaufmann, San ~Mateo, Calif., pp. 290-294.Google Scholar
- ~FIEDLER, M. 1972. Bounds for eigenvalues of doubly stochastic matrices. Lin.4lg Appl. 5, 3 ~(July), 299-310.Google Scholar
- ~F1LL, J. A. 1991. Eigenvalue bounds on convergence to stationarity for nonreversible Markov ~ chains, with an apphcation to the exclusion process. Ann Applied Prob. l, 1, 62-87.Google Scholar
- ~FRANKLIN, J. N. 1968. Matrix Theory. Prentice-Hall, Englewood Cliffs, N.J.Google Scholar
- ~GOLD, E. M. 1967. Language identification in the limit, h~f. Cont. 10, 447-474.Google Scholar
- ~GOLD, E. M. 1972. System identification via state characterization. Autornattca, 8, 621 636.Google Scholar
- ~GOLD, E. M. 1978. Complexity of automaton identification from given data. Inf. Cont. 37, ~302-320.Google Scholar
- ~HARTMANIS, J., AND STEARNS, R. E. 1966. Algebrazc Structure TheoO, of Sequential Machines. ~Prentice-Hall, Englewood Cliffs, N. J. Google Scholar
- ~KEARNS, M., AND VALIANT, L. G. 1994. Cryptographm limitations on learning Boolean formulae ~and finite automata. J. ACM 41, 1 (Jan), 67-95. Google Scholar
- ~KOHAVL Z. 1978. Swztchzng and Fintte Automata Theoo,. McGraw-Hill, New York.Google Scholar
- ~LOV/SZ, L. 1979. Combbmtorial Problems and Exercises. North-Holland, Amsterdam, The ~Netherlands.Google Scholar
- ~PITT, L. 1989. Inductive inference, DFAs, and computational complexity. In Plvceedings of the ~1989 blternational Workshop on Analogical and bzductiue blference. Lecture Notes in Computer ~Science. Springer-Verlag, New York. Google Scholar
- ~PITT, t., AND WARMUTH, M. K. 1993. The minimum consistent DFA problem cannot be ~approximated within any polynomial. J. ACM 40, 1 (Jan.), 95-142. Google Scholar
- ~PITT, t., AND WARMUTH, M. K. 1990. Prediction-preserving reducibility. J. Comput. Syst. Sci. 41, ~3 (Dec.), 430-467. Google Scholar
- ~RIVEST R. L., AND SCHAPIRE, R. E. 1989. Inference of finite automata using homing sequences. ~In Proceedings of the 21st Annual ACM Symposium on Theoly of Computing, Inf. Comput. ACM, ~New York. Also blf. Comput. 103 (Apr.), 299-347. Google Scholar
- ~ DE SAINT-EXUPERY, A. 1943. The Little Prfi~ce. Harcourt, Brace, & World, New York.Google Scholar
- ~ SCHAPIRE, R. E. 1992. The Design and Analysis of Efficient Learning Algorithms. MIT Press, ~Cambridge, Mass. Google Scholar
- ~TARJAN, R. E. 1975. Efficiency of a good but not linear set union algorithm. J. ACM 22, 2 (Apr), ~215-225. Google Scholar
- ~TRAKHTENBROT, B. A., AND BARZDIN', YA. M. 1973. Finite Automata: Behacior and Synthesis. ~North-Holland, Amsterdam, The Netherlands.Google Scholar
Index Terms
Diversity-based inference of finite automata
Recommendations
Nondeterminism versus determinism for two-way finite automata: generalizations of Sipser's separation
ICALP'03: Proceedings of the 30th international conference on Automata, languages and programmingWhether there exists an exponential gap between the size of a minimal deterministic two-way automaton and the size of a minimal nondeterministic two-way automaton for a specific regular language is a long standing open problem and surely one of the most ...
On the power of quantum finite state automata
FOCS '97: Proceedings of the 38th Annual Symposium on Foundations of Computer ScienceIn this paper, we introduce 1-way and 2-way quantum finite state automata (1qfa's and 2qfa's), which are the quantum analogues of deterministic, nondeterministic and probabilistic 1-way and 2-way finite state automata. We prove the following facts ...
Descriptional complexity of determinization and complementation for finite automata
CATS 2011: Proceedings of the Seventeenth Computing on The Australasian Theory Symposium - Volume 119In this paper we study the subset construction that transforms nondeterministic finite automata (NFA) to deterministic finite automata (DFA). It is well known that given a n-state NFA, the subset construction algorithm produces a 2n-state DFA in the ...






Comments