Error-Correcting Factorization

摘要: Error Correcting Output Codes (ECOC) is a successful technique in multi-class classification, which core problem Pattern Recognition and Machine Learning. A major advantage of ECOC over other methods that the multi- class decoupled into set binary problems are solved independently. However, literature defines general error-correcting capability for ECOCs without analyzing how it distributes among classes, hindering deeper analysis pair-wise error-correction. To address these limitations this paper proposes an Error-Correcting Factorization (ECF) method, our contribution three fold: (I) We propose novel representation error-correction capability, called design matrix, enables us to build on basis allocating correction pairs classes. (II) derive optimal code length using rank properties matrix. (III) ECF formulated as discrete optimization problem, relaxed solution found efficient constrained block coordinate descent approach. (IV) Enabled by flexibility introduced with matrix we allocate classes prone confusion. Experimental results several databases show when confusable outperforms state-of-the-art approaches.