knnsearch(trainingset, testingset, 'K', neighborsvalue) on GPU matlab using parallel processing

how to apply knnsearch builtin function of searching the neighbors in GPU using parallel processing , having problem in knnsearch function with Cuda.

Answer 1

[Joss Knight]

arrayfun is for applying a scalar operation element-wise across an array. You should not be using it to wrap knnsearch. Just call knnsearch with your gpuArray inputs.

See help gpuArray\knnsearch.

[Dang Manh Truong]

Are you sure? knnsearch is not on the list of function that supports gpuArray, and I tried but it crashed :(

[Joss Knight]

I'm sure! Probably you don't have R2016b.

http://www.mathworks.com/help/stats/release-notes.html

[Hao Zhang]

But it seems the GPU version of knnsearch only support brute force method, for large data set, it is slower than the CPU version with kdtree algorithm.

Answer 2

[KSSV]

https://devtalk.nvidia.com/default/topic/412423/fast-k-nearest-neighbor-search-using-gpu-code-now-available/

http://nichol.as/papers/Garcia/Fast%20k%20nearest%20neighbor%20search%20using.pdf