A formula is read-once if each variable appears at most once in it. An arithmetic read-once formula is one in which the operators are addition, subtraction, multiplication, and division. We present polynomial time algorithms for exact learning of arithmetic read-once formulas over a field. We present a membership and equivalence query algorithm that identifies arithmetic read-once formulas over an arbitrary field. We present a randomized membership query algorithm (i.e., a randomized black box interpolation algorithm) that identifies such formulas over finite fields with at least $2n + 5$ elements (where n is the number of variables) and over infinite fields. We also show the existence of nonuniform deterministic membership query algorithms for arbitrary read-once formulas over fields of characteristic 0, and division-free read-once formulas over fields that have at least $2n^{3} + 1$ elements. For our algorithms, we assume we are able to perform efficiently arithmetic operations on field elements and compute square roots in the field. It is shown that the ability to compute square roots is necessary in the sense that the problem of computing $n - 1$ square roots in a field can be reduced to the problem of identifying an arithmetic formula over n variables in that field. Our equivalence queries are of a slightly nonstandard form, in which counterexamples are required not to be inputs on which the formula evaluates to $0/0$. This assumption is shown to be necessary for fields of size $o(n/ \log n)$ in the sense that we prove there exists no polynomial time identification algorithm that uses only membership and standard equivalence queries.

A number of efficient learning algorithms achieve exact identification of an unknown function from some class using membership and equivalence queries. Using a standard transformation such algorithms can easily be converted to on-line learning algorithms that use membership queries. Under such a transformation the number of equivalence queries made by the query algorithm directly corresponds to the number of mistakes made by the on-line algorithm. In this paper we consider several of the natural classes known to be learnable in this setting, and investigate the minimum number of equivalence queries with accompanying counterexamples (or equivalently the minimum number of mistakes in the on-line model) that can be made by a learning algorithm that makes a polynomial number of membership queries and uses polynomial computation time. We are able both to reduce the number of equivalence queries used by the previous algorithms and often to prove matching lower bounds. As an example, consider...

We present a polynomial time algorithm that PAC learns kp decision trees on the uniform distribution for an arbitrary constant k (a p decision tree tests each attribute on a single node, and a kp decision tree tests each attribute on at most k different nodes).The algorithm produces a hypothesis that is a decision tree (t bough not necessarily k~).The algorithm haa the flavor of empirical decision tree algorithms in that it builds a hypothesis by choosing variables on which to split and then recursively solves smaller problems without backtracking.1

A formula is read-once if each variable appears on at most a single input. Angluin, Hellerstein, and Karpinski have shown that boolean formulas with AND, OR, and NOT gates are exactly identifiable in polynomial time using membership and equivalence queries [AHK89]. Hancock and Hellerstein have generalized this to allow a wider subclass of symmetric basis functions [HH91]. We show a polynomial time algorithm in this model for identifying read-once formulas whose gates compute arbitrary functions of fan-in k or less for some constant k (i.e. any f :{0,1}1≤c≤k → {0,1}). We further show that if there is a polynomial time membership and equivalence query algorithm to identify read-once formulas over some set of functions B that meets certain technical conditions, then there is also such an algorithm to identify read-once formulas over Bu{f:{0,1}1≤c≤k → {0,1}}. Finally, we extend the previous results to show that there is a polynomial time identification algorithm for read-once formulas over the basis of all symmetric functions (and hence also over the union of arbitrary symmetric and arbitrary constant fan-in gates). Given standard cryptographic assumptions, none of these results are possible for read-twice formulas.

A formula is read-once if each variable appears at most once in it. An arithmetic read-once formula is one in which the operators are addition, subtraction, multiplication, and division. We present polynomial time algorithm for exactly learning (or interpolating) arithmetic read-once formulas computing functions over a field. We present an algorithm that uses randomized membership queries (or substitutions) to identify such formulas over large finite fields and infinite fields. We also present a deterministic algorithm that uses equivalence queries as well as membership queries to identify arithmetic read-once formulas over small finite fields. We then non-constructively show the existence of deterministic membership query (interpolation) algorithms for arbitrary formulas over fields of characteristic 0 and for division-free formulas over large or infinite fields. Our algorithms assume we are able to efficiently perform arithmetic operations on field elements and compute square roots in the field. It is shown that the ability to compute square roots is necessary, in the sense that the problem of computing n – 1 square roots in a field can be reduced to the problem of identifying an arithmetic formula over n variables in that field. Our equivalence queries are of a slightly non-standard form, in which counterexamples are required to not be inputs on which the formula evaluates to 0/0. This assumption is shown to be necessary for fields of size o(n/log n), for which it is shown that there is no polynomial time identification algorithm that uses just membership and standard equivalence queries.