Mingfei Huang: Accelerating AI from Cloud to Edge
summary
An open-source edge AI stack that targets inference only and runs on Arm SoCs with heterogeneous CPU+GPU+DL-accelerator+DSP, packaged as OpenAILab platform with three layers: compute libraries, a framework layer and a domain API layer. It uses an optimized Caffe HRT and a custom Tianjin inference engine with dynamic scheduler that picks between ACL/OpenBLAS/HCL kernels, offers debugging/profiling and quantization tools, ships vision and speech models (face, MTCNN, keyword speech) on GitHub, and runs on HiKey960, DragonBoard and Rock960.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:41OpenAI Lab
- 1:35General Idea
- 9:17Optimized Cafe
- 14:37Inference Engine
- 19:08Domain Library
- 22:54Open Platform
- 23:34Github
related talks
09mTackling edge computing with Intel's AI Builders program
37mEnabling AI For Everyone - Hiai Open Platform Introduction
31mInfiniEdge AI: Bringing Open-Source Generative AI To the Edge
26mBuilding a Cloud Native Edge Platform for Accelerating AI Inference
48mThe Power of Edge AI: Revolutionizing Future of Real-Time, Autonomous Processing
10mThe Rise and Impact of Edge AI
37mDeep Learning Neural Network Acceleration at the Edge
36mApplied AI on the edge