A Provable Approach for Double-Sparse Coding, AAAI-2018

The arXiv paper is on https://arxiv.org/abs/1711.03638

Goal

Learn to recover the sparse dictionary $A^*$ from samples $y$ generated by a generative model $y = A^*x + \varepsilon$.

Data and simulation setting

• Deterministic, sparse, orthonormal dictionary $A^*$
• Sparse, random and overcomplete dictionary

Required Matlab packages and/or libraries:

• gaimc for the bipartite_matchings algorithm
• To compare our algorithm with Trainlets, you need to download Trainlets (OSDL) code and put it inside this folder. Note that some libs (mtimesx, omps and so on) are required to run this program.

How to run

Set the running mode for algorithm you want to test and run run_simulation.m.

Contact

[Thanh Nguyen](thanhng at iastate dot edu)