2018/02/11 by Bernd Malle, Malle, Bernd, Nicola Giuliani +5
Computer Science · #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.1802.03707
AI applications pose increasing demands on performance, so it is not\nsurprising that the era of client-side distributed software is becoming\nimportant. On top of many AI applications already using mobile hardware, and\neven browsers for computationally demanding AI applications, we are already\nwitnessing the emergence of client-side (federated) machine learning\nalgorithms, driven by the interests of large corporations and startups alike.\nApart from mathematical and algorithmic concerns, this trend especially demands\nnew levels of computational efficiency from client environments. Consequently,\nthis paper deals with the question of state-of-the-art performance by\npresenting a comparison study between native code and different browser-based\nimplementations: JavaScript, ASM.js as well as WebAssembly on a representative\nmix of algorithms. Our results show that current efforts in runtime\noptimization push the boundaries well towards (and even beyond) native binary\nperformance. We analyze the results obtained and speculate on the reasons\nbehind some surprises, rounding the paper off by outlining future possibilities\nas well as some of our own research efforts.\n