Machine learning frameworks such as TensorFlow require large numbers of vector and matrix calculations during both training and inference. The performance and power consumption of machine learning applications can be vastly improved by using parallel computing but how can this be achieved on mobile devices? While a computation expressed using TensorFlow can be executed across heterogeneous systems, support has so far been limited to NVIDIA ® processors using CUDA ® making it difficult to target embedded and mobile hardware. Using SYCL™, developers can use OpenCL devices and write applications that execute across heterogeneous platforms.
Required audience experience: Experience of writing machine learning applications required.
Objective of the talk: Attendees will learn about the challenges of targeting embedded and mobile hardware, what solutions are available so that training and inference can be performed on the device rather than in the cloud, and what OpenCL is and how it can be used to target a range of hardware from a single code base.
Keywords: OpenCL, TensorFlow
You can view Rod’s slides via the link below:
Rod Burns: Building Machine Learning Applications for Mobile and Embedded Devices
You can watch Rod’s presentation below: