Open Source AI Infrastructure and Ecosystem

Deep neural network (DNN) learning driven by big data has become the state-of-the-art system for image/speech recognition, sequence learning and reinforcement learning. The remarkable progress in AI has been made possible by the availability of: 1) huge amount of labelled training data, 2) modern compute capability (e.g., GPU, TPU or Huawei A910), and 3) broadly accessible open source software (e.g., Kubernetes/Docker, Hadoop/Spark, TensorFlow/PyTorch/CNTK). In this talk, we will discuss open source AI infrastructure and ecosystem. We are in the process of building and deploying an open source AI infrastructure and ecosystem that is agnostic of underlying hardware and AI framework. With the help of our AI infrastructure, scientists/researchers can quickly build research environment and system, reproduce the state-of-the-art AI result, share and track colleague’s work, conduct sophisticated research and development (with state-of-the art AI tool such as training on a distributed machine cluster with multiple GPU/NPU, hyperparameter tuning on a distributed cluster), and deployed the developed model into a production ready system.
Who is this presentation for?
AI scientist and engineer.
Prerequisite knowledge:
N/A
What you'll learn?

Dr. Jin Li is the CTO and Co-founder of Apulis Technology Inc. Prior to founding Apulis, he was a Partner Research Manager of Cloud Computing and Storage Group in Microsoft Research (Redmond, US). His work on Local Reconstruction Code, data deduplication and high performance SSD storage has been widely used in Azure and Microsoft Server line, and lead to hundred of million dollars saving per annum, won a Microsoft TCN storage achievement award, and a 2012 Usenix ATC best paper award. His work on DLWorkspace has become the main AI training platform for Bing and Microsoft Cognitivie Service. Most of leading deep learning work of Microsoft has been developed up the platform. Dr. Li is an IEEE Fellow。