ar X iv : 0 81 0 . 23 11 v 1 [ cs . A I ] 1 3 O ct 2 00 8 Non - Negative Matrix Factorization , Convexity and Isometry ∗
نویسندگان
چکیده
In this paper we explore avenues for improving the reliability of dimensionality reduction methods such as Non-Negative Matrix Factorization (NMF) as interpretive exploratory data analysis tools. We first explore the difficulties of the optimization problem underlying NMF, showing for the first time that non-trivial NMF solutions always exist and that the optimization problem is actually convex, by using the theory of Completely Positive Factorization. We subsequently explore four novel approaches to finding globallyoptimal NMF solutions using various ideas from convex optimization. We then develop a new method, isometric NMF (isoNMF), which preserves non-negativity while also providing an isometric embedding, simultaneously achieving two properties which are helpful for interpretation. Though it results in a more difficult optimization problem, we show experimentally that the resulting method is scalable and even achieves more compact spectra than standard NMF.
منابع مشابه
ar X iv : 0 81 0 . 46 11 v 1 [ cs . L G ] 25 O ct 2 00 8 Learning Isometric Separation Maps ∗
Maximum Variance Unfolding (MVU) and its variants have been very successful in embedding data-manifolds in lower dimensionality spaces, often revealing the true intrinsic dimensions. In this paper we show how to also incorporate supervised class information into an MVU-like method without breaking its convexity. We call this method the Isometric Separation Map and we show that the resulting ker...
متن کاملar X iv : 0 71 0 . 35 19 v 1 [ cs . C C ] 1 8 O ct 2 00 7 P - matrix recognition is co - NP - complete
This is a summary of the proof by G.E. Coxson [1] that P-matrix recognition is co-NP-complete. The result follows by a reduction from the MAX CUT problem using results of S. Poljak and J. Rohn [5].
متن کاملar X iv : h ep - l at / 0 11 00 06 v 2 3 O ct 2 00 1 1 Matrix elements of ∆ S = 2 operators with Wilson fermions
متن کامل
ar X iv : 0 81 0 . 50 56 v 1 [ cs . C C ] 2 8 O ct 2 00 8 P is not equal to NP . Sten - Åke
SAT 6∈ P is true, and provable in a simply consistent extension B of a first order theory B of computing, with a single finite axiom B characterizing a universal Turing machine. Therefore P 6 = NP is true, and provable in a simply consistent extension B of B.
متن کامل