TY - JOUR

T1 - Conjunctions of Unate DNF Formulas

T2 - Learning and Structure

AU - Feigelson, Aaron

AU - Hellerstein, Lisa

N1 - Funding Information:
* Research partially funded by NSF Grants CCR-95-01660 and CCR-92-10957. -Current address: Dept. of Computer and Information Science, Polytechnic University, 5 Metrotech Center, Brooklyn, New York 11201.

PY - 1998/2/1

Y1 - 1998/2/1

N2 - A central topic in query learning is to determine which classes of Boolean formulas are efficiently learnable with membership and equivalence queries. We consider the class ℛk consisting of conjunctions of k unate DNF formulas. This class generalizes the class of k-clause CNF formulas and the class of unate DNF formulas, both of which are known to be learnable in polynomial time with membership and equivalence queries. We prove that ℛ2 can be properly learned with a polynomial number of polynomial-size membership and equivalence queries, but can be properly learned in polynomial time with such queries if and only if P = NP. Thus the barrier to properly learning ℛ2 with membership and equivalence queries is computational rather than informational. Few results of this type are known. In our proofs, we use recent results of Hellerstein et al. (1997, J. Assoc. Comput. Mach. 43 (5), 840-862), characterizing the classes that are polynomial-query learnable, together with work of Bshouty on the monotone dimension of Boolean functions. We extend some of our results to ℛ2 and pose open questions on learning DNF formulas of small monotone dimension. We also prove structural results for ℛk. We construct, for any fixed k ≥ 2, a class of functions f that cannot be represented by any formula in ℛk, but which cannot be "easily" shown to have this property. More precisely, for any function f on n variables in the class, the value of f on any polynomial-size set of points in its domain is not a witness that f cannot be represented by a formula in ℛk. Our construction is based on BCH codes.

AB - A central topic in query learning is to determine which classes of Boolean formulas are efficiently learnable with membership and equivalence queries. We consider the class ℛk consisting of conjunctions of k unate DNF formulas. This class generalizes the class of k-clause CNF formulas and the class of unate DNF formulas, both of which are known to be learnable in polynomial time with membership and equivalence queries. We prove that ℛ2 can be properly learned with a polynomial number of polynomial-size membership and equivalence queries, but can be properly learned in polynomial time with such queries if and only if P = NP. Thus the barrier to properly learning ℛ2 with membership and equivalence queries is computational rather than informational. Few results of this type are known. In our proofs, we use recent results of Hellerstein et al. (1997, J. Assoc. Comput. Mach. 43 (5), 840-862), characterizing the classes that are polynomial-query learnable, together with work of Bshouty on the monotone dimension of Boolean functions. We extend some of our results to ℛ2 and pose open questions on learning DNF formulas of small monotone dimension. We also prove structural results for ℛk. We construct, for any fixed k ≥ 2, a class of functions f that cannot be represented by any formula in ℛk, but which cannot be "easily" shown to have this property. More precisely, for any function f on n variables in the class, the value of f on any polynomial-size set of points in its domain is not a witness that f cannot be represented by a formula in ℛk. Our construction is based on BCH codes.

UR - http://www.scopus.com/inward/record.url?scp=0346968409&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0346968409&partnerID=8YFLogxK

U2 - 10.1006/inco.1997.2684

DO - 10.1006/inco.1997.2684

M3 - Article

AN - SCOPUS:0346968409

SN - 0890-5401

VL - 140

SP - 203

EP - 228

JO - Information and Computation

JF - Information and Computation

IS - 2

ER -