Nonnegative Signal Decomposition with Supervision

Abstract EN

This paper presents a novel algorithm to numerically decompose mixed signals in a collaborativeway, given supervision of the labels that each signal contains.

The decomposition is formulated as anoptimization problem incorporating nonnegative constraint.

A nonnegative data factorization solution ispresented to yield the decomposed results.

It is shown that the optimization is efficient and decreases theobjective function monotonically.

Such a decomposition algorithm can be applied on multilabel trainingsamples for pattern classification.

The real-data experimental results show that the proposed algorithmcan significantly facilitate the multilabel image classification performance with weak supervision.