A Clustering Approach to Learning Sparsely Used Overcomplete Dictionaries

作者：Agarwal Alekh^*; Anandkumar Animashree; Netrapalli Praneeth

来源：IEEE Transactions on Information Theory, 2017, 63(1): 575-592.

DOI：10.1109/TIT.2016.2614684

• Dictionary learning
• sparse coding
• overcomplete dictionaries
• incoherence
• lasso

We consider the problem of learning overcomplete dictionaries in the context of sparse coding, where each sample selects a sparse subset of dictionary elements. Our main result is a strategy to approximately recover the unknown dictionary using an efficient algorithm. Our algorithm is a clustering-style procedure, where each cluster is used to estimate a dictionary element. The resulting solution can often be further cleaned up to obtain a high accuracy estimate, and we provide one simple scenario where l(1)-regularized regression can be used for such a second stage.

• 出版日期2017-1
• 单位Microsoft