Solution of linear ill-posed problems using overcomplete dictionaries

TL;DR

This paper introduces a novel Lasso-based method for solving ill-posed linear inverse problems using overcomplete dictionaries, avoiding restrictive assumptions and demonstrating effective risk bounds and computational performance.

Contribution

It proposes an innovative approach that inverts dictionary functions to apply Lasso without compatibility conditions, extending to mixture density estimation.

Findings

Provides oracle inequalities for finite sample risk

Demonstrates good computational properties through simulations

Extends methodology to mixture density estimation

Abstract

In the present paper we consider application of overcomplete dictionaries to solution of general ill-posed linear inverse problems. Construction of an adaptive optimal solution for such problems usually relies either on a singular value decomposition or representation of the solution via an orthonormal basis. The shortcoming of both approaches lies in the fact that, in many situations, neither the eigenbasis of the linear operator nor a standard orthonormal basis constitutes an appropriate collection of functions for sparse representation of the unknown function. In the context of regression problems, there have been an enormous amount of effort to recover an unknown function using an overcomplete dictionary. One of the most popular methods, Lasso, is based on minimizing the empirical likelihood and requires stringent assumptions on the dictionary, the, so called, compatibility conditions. While these conditions may be satisfied for the original dictionary functions, they usually do not hold for their images due to contraction imposed by the linear operator. In what follows, we bypass this difficulty by a novel approach which is based on inverting each of the dictionary functions and matching the resulting expansion to the true function, thus, avoiding unrealistic assumptions on the dictionary and using Lasso in a predictive setting. We examine both the white noise and the observational model formulations and also discuss how exact inverse images of the dictionary functions can be replaced by their approximate counterparts. Furthermore, we show how the suggested methodology can be extended to the problem of estimation of a mixing density in a continuous mixture. For all the situations listed above, we provide the oracle inequalities for the risk in a finite sample setting. Simulation studies confirm good computational properties of the Lasso-based technique.