Programming Research Group A LEARNABILITY MODEL FOR UNIVERSAL REPRESENTATIONS
نویسندگان
چکیده
This paper deenes a new computational model of inductive learning, called U-learnability (Universal Learnability), that is well-suited for rich representation languages, most notably for universal (Turing equivalent) representations. It is motivated by three observations. Firstly, existing computational models of inductive learning|the best known of which are identiication in the limit and PAC-learnability|either fail to provide guarantees of accuracy and computational eeciency or require severely restricted representation languages. Secondly, practical machine learning programs use rich representations yet often perform eeciently and achieve high degrees of accuracy. Thirdly, these machine learning programs often make assumptions about the underlying probability distribution over target concepts; such an assumption provides a soft bias for learning. This contrasts with existing computational models of inductive learning, which incorporate only language biases, or hard biases. U-learnability incorporates soft biases in the form of a Bayesian prior distribution over target concepts. (Other authors have argued for the use of Bayesian prior distributions in a manner more restricted than their use in U-learnability, though this has not lead to a new model of polynomial-time learnability.) In addition, due to problems of undecidability with universal representations, U-learnability makes special provision for the use of time-bounded concepts. A time bound takes the form of a polynomial function of the size of an example.
منابع مشابه
A Learnability Model for Universal Representations and Its Application to Top-down Induction of Decision Trees
Automated inductive learning is a vital part of machine intelligence and the design of intelligent agents. A useful formalization of inductive learning is the model of PAC-learnability. Nevertheless, the ability to learn every target concept expressible in a given representation, as required in the PAC-learnability model, is highly demanding and leads to many negative results for interesting co...
متن کاملA Universal Investigation of $n$-representations of $n$-quivers
noindent We have two goals in this paper. First, we investigate and construct cofree coalgebras over $n$-representations of quivers, limits and colimits of $n$-representations of quivers, and limits and colimits of coalgebras in the monoidal categories of $n$-representations of quivers. Second, for any given quivers $mathit{Q}_1$,$mathit{Q}_2$,..., $mathit{Q}_n$, we construct a new quiver $math...
متن کاملDeformation of Outer Representations of Galois Group II
This paper is devoted to deformation theory of "anabelian" representations of the absolute Galois group landing in outer automorphism group of the algebraic fundamental group of a hyperbolic smooth curve defined over a number-field. In the first part of this paper, we obtained several universal deformations for Lie-algebra versions of the above representation using the Schlessinger criteria for...
متن کاملUniversal Central Extension of Current Superalgebras
Representation as well as central extension are two of the most important concepts in the theory of Lie (super)algebras. Apart from the interest of mathematicians, the attention of physicist are also drawn to these two subjects because of the significant amount of their applications in Physics. In fact for physicists, the study of projective representations of Lie (super)algebras are very impo...
متن کاملModeling the acquisition of quantifier semantics: a case study in function word learnability
This paper studies the acquisition of quantifier meanings as a case study of function word learnability. We suggest that learners construct semantic representations of quantifiers and other function words using a compositional statistical learning mechanism that operates over a small set of domaingeneral cognitive primitives. We present a simple cross-situational learning model that provably so...
متن کامل