PAC - learning and Asymptotic System Identi cationTheory
نویسنده
چکیده
In this paper we discuss PAC-learning of functions from a traditional System Identiicaton perspective. The well established asymptotic theory for the iden-tiied models' properties is reviewed from the PAC-learning perspective. The role of nite-dimensional, smooth parametrizations over compact parameter sets is spelled out. This also sets some limits for the interest of identiication-theory type results in a learning-tehory context.
منابع مشابه
Identi cation Criteria and Lower Bounds for Perceptron-like Learning Rules
Perceptron-like learning rules are known to require exponentially many correction steps in order to identify Boolean threshold functions exactly. We introduce criteria that are weaker than exact identiication and investigate whether learning becomes signiicantly faster if exact identiication is replaced by one of these criteria: PAC identiication, order identiication, and sign iden-tiication. P...
متن کاملSystem Identi cation: A Learning Theory Approach
In this paper, the problem of system identi cation is formulated as a problem in statistical learning theory. The main motivation for such a reformulation is that traditionally system identi cation theory provides asymptotic results, whereas statistical learning theory is devoted to the derivation of nite time estimates. If system identi cation is to be combined with robust control theory to de...
متن کاملLearning dynamical systems in a stationary environment
We consider the problem of learning the input–output relation of a dynamical system from noisy data. Our method rests on the use of a smooth simultaneous estimator which generalizes the standard empirical estimator. In a stationary environment, our algorithm is shown to select a model which exhibits the Probably Approximately Correct (PAC) property under very mild conditions. This contribution ...
متن کاملLearning DFA for Simple Examples
We present a framework for learning DFA from simple examples. We show that e cient PAC learning of DFA is possible if the class of distributions is restricted to simple distributions where a teacher might choose examples based on the knowledge of the target concept. This answers an open research question posed in Pitt's seminal paper: Are DFA's PAC-identi able if examples are drawn from the uni...
متن کاملAgnostic PAC-Learning of Functions on Analog Neural Nets
There exist a number of negative results ([J), [BR), [KV]) about learning on neural nets in Valiant's model [V) for probably approximately correct learning ("PAC-learning"). These negative results are based on an asymptotic analysis where one lets the number of nodes in the neural net go to infinit.y. Hence this analysis is less adequate for the investigation of learning on a small fixed neural...
متن کامل