Learnability of Relatively Quantified Generalized Formulas

In this paper we study the learning complexity of a vast class of quantifed formulas called Relatively Quantified Generalized Formulas. This class of formulas is parameterized by a set of predicates, called a basis. We give a complete classification theorem, showing that every basis gives rise to quantified formulas that are either polynomially learnable with equivalence queries, or not polynomially predictable with membership queries under some cryptographic assumption. We also provide a simple criteria distinguishing the learnable cases from the non-learnable cases.

[1]  Víctor Dalmau,et al.  A Dichotomy Theorem for Learning Quantified Boolean Formulas , 1997, COLT.

[2]  David P. Williamson,et al.  A complete classification of the approximability of maximization problems derived from Boolean constraint satisfaction , 1997, STOC '97.

[3]  Luca Trevisan,et al.  Constraint satisfaction: the approximability of minimization problems , 1997, Proceedings of Computational Complexity. Twelfth Annual IEEE Conference.

[4]  Ágnes Szendrei,et al.  Clones in universal algebra , 1986 .

[5]  D. Angluin.  Queries and Concept Learning , 1988 .

[6]  Peter Jeavons,et al.  Quantified Constraints: Algorithms and Complexity , 2003, CSL.

[7]  Dana Angluin,et al.  When won't membership queries help? , 1991, STOC '91.

[8]  Leslie G. Valiant,et al.  Cryptographic limitations on learning Boolean formulae and finite automata , 1994, JACM.

[9]  Nader H. Bshouty.  Exact Learning Boolean Function via the Monotone Theory , 1995, Inf. Comput..

[10]  Leslie G. Valiant,et al.  A theory of the learnable , 1984, STOC '84.

[11]  Martin C. Cooper,et al.  Constraints, Consistency and Closure , 1998, Artif. Intell..

[12]  Thomas J. Schaefer,et al.  The complexity of satisfiability problems , 1978, STOC.

[13]  Nicholas Pippenger,et al.  Theories of computability , 1997 .

[14]  Nadia Creignou,et al.  A Dichotomy Theorem for Maximum Generalized Satisfiability Problems , 1995, J. Comput. Syst. Sci..

[15]  Nader H. Bshouty,et al.  Exploring Learnability between Exact and PAC , 2002, COLT.

[16]  Peter Jeavons,et al.  Learnability of quantified formulas , 1999, Theor. Comput. Sci..

[17]  Hubie Chen.  Quantified Constraint Satisfaction Problems: Closure Properties, Complexity, and Proof Systems , 2003 .

[18]  Michael Frazier,et al.  Learning conjunctions of Horn clauses , 2004, Machine Learning.

[19]  Andrei A. Bulatov,et al.  Tractable conservative constraint satisfaction problems , 2003, 18th Annual IEEE Symposium of Logic in Computer Science, 2003. Proceedings..

[20]  Nadia Creignou,et al.  Complexity of Generalized Satisfiability Counting Problems , 1996, Inf. Comput..

[21]  Emil L. Post.  The two-valued iterative systems of mathematical logic , 1942 .

[22]  Jeffrey C. Jackson.  An Efficient Membership-Query Algorithm for Learning DNF with Respect to the Uniform Distribution , 1997, J. Comput. Syst. Sci..

[23]  Sanjeev Khanna,et al.  Complexity classifications of Boolean constraint satisfaction problems , 2001, SIAM monographs on discrete mathematics and applications.