govg / podcuda

POD library on CUDA

##PODCUDA

This library aims to be a CUDA based implementation of the POD.

###POD (Proper Orthogonal Decomposition)

POD or Proper Orthogonal Decomposition is a tool used in data analysis.

In simple terms, it works by generating a lower dimensional approximate of noisy high dimensional data. It is closely related to PCA, SVD and a lot of other such methods

###CUDA

CUDA or Compute Unified Device Architecture, is a parallel computing platform created by NVIDIA corp. It allows using general GPUs present in mid-high end computers for computation. The advantage of using this over traditional CPU based computation is that there are generally around hundreds of compute cores on a normal GPU. While current generation of GPUs do not allow better than single (double in case of high end) precision computation, it still is good enough for a lot of trivially parallelizable computation.

---

The primary focus of this library will be ease of use, and speed. CUDA does have its limitations, and I will try to work around it as much as it allows.

The target is to be finished with single precision computation in the coming month, so that work on double precision versions can be started by the start of October.

---

###Related work and references :

Related decompositions :

• Eigen Decomposition
• K-L Theorem

Helpful tutorials and introductions :

• http://web.stanford.edu/~amsallem/CA-CME345-Ch4.pdf
• http://www.iisc.ernet.in/currsci/apr102000/tutorial2.pdf

For queries, contact me at [e-mail].

I also lurk on Quakenet and Freenode, govg in both.