From much of the previous posts, we find that matrix factorization has many other uses besides haunting college students, such as in RSA security, sensors, and data mining. Yet another application of matrix factorization is in protein folding recognition. Protein folding is a very critical biological process that remains largely a mystery, requiring much computational power to fully understand. Therefore breakthroughs in this discovery process involving efficient matrix factorizations are extremely significant.

Researchers have not only found methods of matrix factorizations in protein folding, but have also found a much faster algorithm for non-negative matrix factorizations (NMF). This complicated algorithm modifies the traditional NMF in two ways: scaling (normalization) and the addition of a small random number to each element in one of the factorized matrices. The details of this method and its proof occurs in part 2 of the document linked below.

When actually applying data to this new process, researchers have found that if correctly implemented, the new NMF algorithm converges 11 times faster than the original. At the same time, this algorithm also achieves a much higher accuracy. If you would like to know more, follow this link, click on “Pdf”, and log in using:
Web Account: cs322blog
Password: dougjames