Customising parallelism and caching for machine learning | IEEE Conference Publication | IEEE Xplore