Attribute-Efficient Learning

Keywords and Synonyms

Learning with irrelevant attributes            

Problem Definition

Given here is a basic formulation using the online mistake bound model, which was used by Littlestone [9] in his seminal work.

Fix a class C of Boolean functions over n variables. To start a learning scenario, a target function ${ f_\ast\in C }$ is chosen but not revealed to the learning algorithm. Learning then proceeds in a sequence of trials. At trial t, an input ${ \boldsymbol{x}_t \in \{0,1 \}^n }$ is first given to the learning algorithm. The learning algorithm then produces its prediction ${ \hat{y}_t }$, which is its guess as to the unknown value ${ f_\ast(x_t) }$. The correct value ${ y_t=f_\ast(x_t) }$ is then revealed to the learner. If ${ y_t\neq\hat{y}_t }$, the learning algorithm made a mistake. The learning algorithm learns C with mistake bound m, if the number of mistakes never exceeds m, no matter how many trials are made and how f _∗ and \( {...