English

Unsupervised model-free representation learning

Machine Learning 2020-05-05 v4 Quantitative Methods Machine Learning

Abstract

Numerous control and learning problems face the situation where sequences of high-dimensional highly dependent data are available but no or little feedback is provided to the learner, which makes any inference rather challenging. To address this challenge, we formulate the following problem. Given a series of observations X0,,XnX_0,\dots,X_n coming from a large (high-dimensional) space X\mathcal X, find a representation function ff mapping X\mathcal X to a finite space Y\mathcal Y such that the series f(X0),,f(Xn)f(X_0),\dots,f(X_n) preserves as much information as possible about the original time-series dependence in X0,,XnX_0,\dots,X_n. We show that, for stationary time series, the function ff can be selected as the one maximizing a certain information criterion that we call time-series information. Some properties of this functions are investigated, including its uniqueness and consistency of its empirical estimates. Implications for the problem of optimal control are presented.

Cite

@article{arxiv.1304.4806,
  title  = {Unsupervised model-free representation learning},
  author = {Daniil Ryabko},
  journal= {arXiv preprint arXiv:1304.4806},
  year   = {2020}
}

Comments

The update is the journal version appearing in IEEE IT transactions under the title "Time-series information and unsupervised learning of representations." This version includes important corrections and new results. Some of the results (presented in previous versions) were reported at ISIT'13 and ALT'13

R2 v1 2026-06-22T00:01:36.716Z