Table 21.

Deep Learning in heterogeneous computing systems.

Year	Reference	Application Area	Highlights	Key Findings	Limitations
2015	[16]	Heterogeneous multicore Architecture	A Heterogeneous multi-core architecture to facilitate the high-performance execution of Machine learning algorithms is explored in this research work. The middleware platform called HeteroSpark involves the integration of a GPU accelerator into an already existing CPU-based architecture (Spark) to accelerate computationally intensive machine learning methods. Employing Java virtual machines, data is moved between the CPU and the GPU when required for intensive computations. The GPU is positioned in Spark’s worker nodes. The GPU relieves the CPU of workload, thus accelerating the entire computation process due to GPUs’ thread-level parallelism. The process is abstracted from the user through user-friendly APIs.	The performance of the proposed heterogeneous framework is compared to a baseline (Spark), and the results were fascinating, with GPU acceleration yielding 18.6x when HeteroSpark is scaled up to “32CPU cores: 8 GPUs” and Spark uses “32 Cores: no GPUs”.	The research focuses on a software approach to the problem and not a hardware perspective.