The classification and regression algorithms are usually implemented to run on processors. Nowadays the general-purpose computing on graphics processing units (GPGPU) implementations are more and more common. In the recent years, the calculation capacities of the GPUs exceed the CPUs’.
My thesis has two main goals. The first is to explore the possibilities using the OpenCL by implementing a popular classification algorithm. On the other hand, it is important to compare the running time of this program with already existing (and commonly used) algorithms. The chosen algorithm is the logistic regression’s batch gradient descent method.
During my work, i use a decent size of training and test data. My main goal is to generate a modell, which has a result what is close to the reference values.
In my thesis, I briefly describe the image classification algorithm from a given image to the label of the image, but the implementation includes only the last step.
The result of my work is a logistic regression batch gradient descent implementation in OpenCL, which not only exceeds the „normal” batch gradient algorithm, but it even faster than the generally faster stochastic gradient descent algorithm.