Training Machine Learning Network on Adreno Mobile GPUs Using OpenCL
Bolan Jiang, Jeng-Hau Lin, Adarsh Golikeri, Li He, Hongqiang Wang, Alex Bourd · 2020
Training on the edge has becoming an interesting topic recently, largely thanks to the inherent drawbacks in cloud-based training methods for mobile users. Traditional methods typically require users to upload their data into cloud, which raises concerns on privacy and data compromise. Users may prefer to have customized models rather than general models from cloud. In addition, it is a waste of resource not to use the computing power of mobile devices.