Subspace Learning with Partial Information
This addresses a data limitation issue in machine learning, but appears incremental as it builds on existing subspace learning methods.
The paper tackles the problem of subspace learning when only partial attributes of instance vectors are observable, proposing efficient algorithms and analyzing their sample complexity.
The goal of subspace learning is to find a $k$-dimensional subspace of $\mathbb{R}^d$, such that the expected squared distance between instance vectors and the subspace is as small as possible. In this paper we study subspace learning in a partial information setting, in which the learner can only observe $r \le d$ attributes from each instance vector. We propose several efficient algorithms for this task, and analyze their sample complexity