Due to open source, the entry threshold of artificial intelligence is lower than ever before. Open source code, frameworks, platforms and tools build a win-win road: it enables the rapid popularization of artificial intelligence, and also enables enterprises participating in open source to receive support and help from the open source community to accelerate their own application research of artificial intelligence. So, what is the key to open source AI?

Artificial intelligence and Big Data are different technical systems. As the cornerstone of digital transformation, they depend on each other in more and more scenarios. Making the two technical systems into one enables many frameworks and tools to work together in one system and give full play to their respective strengths, thus becoming the focus of the industry. This sharing introduces some technical research, application scenarios and practical cases done by ZTE in artificial intelligence and Big Data convergence.

In 2000, Intel publicly released the first open source computer vision library (OpenCV). The release has grown dramatically with millions of downloads per year, and it has become one of the core standards in the industry. In this brand-new deep learning model, Intel has once again created an open source visual inference and neural optimization tool suite(OpenVINO) to solve common problems in the artificial intelligence ecosystem. In many ways, this project is familiar to many of us at Intel. Our evangelist, Mr. Raymond Lo, will tell these unknown stories again, reminding us what points we should pay attention to, so as to contribute to the ultimate success of AI open source products.

Embeddings is the core data form for neural networks to represent semantics. At present, the application of neural network is accelerating to practice. Embeddings, a type of vectorized data, is showing explosive growth. However, at the basic software level, it lacks the ability to manage and analyze vector data. The ZILLIZ team launched the open-source project Milvus to address this issue. Milvus is a vector-based data analysis and search engine, currently incubated in LF AI & DATA Foundation. This report will introduce the Milvus project and share the exploration work of the Milvus community in the process of building a cloud-native vector search engine.

Adlik is an inference optimization tool chain for deep learning model. This speech focuses on the Adlik framework, community operations, and inference acceleration practices.

TinyMS is a high-level API tool developed based on MindSpore. It is an open source deep learning development toolkit, provides an extremely simple high-level API package for the whole process from data preparation to model deployment. Through easy-to-scale modular design, it provides the ability to cover a variety of business scenarios and realizes one-key model training and deployment functions. The lecture mainly shares TinyMS's design architecture, design concept, module analysis, and application practices.

OpenI Open Source Community focuses on building a new generation of artificial intelligence open source community with advanced technology, safety and reliability, rich resources and active atmosphere, forming a long-term ecosystem incentive and operation mechanism adapted to China's characteristics, accumulating and condensing common key technologies of the new generation of artificial intelligence open source projects, and forming a technology and standard system. This sharing introduces the construction experience in recent years around the new generation of artificial intelligence open source basic platform, key technologies, community mechanism and key applications.

In recent years, domestic and foreign mainstream enterprises have released their own deep learning frameworks. Baidu PaddlePaddle, as China's first fully open sourced and functional industrial-level deep learning platform, has been widely practiced and applied inside and outside Baidu. PaddlePaddle has been applied to search, recommendation, advertising and other scenarios on a large scale. At the same time, Baidu has made many world-leading achievements in voice, vision and NLP fields based-on PaddlePaddle Framework. In today's sharing, we will introduce the core capabilities of PaddlePaddle, including ultra-large-scale deep learning model training technology, multi-terminal and multi-platform deployment of high-performance inference engines, and industrial-level open source model libraries and development kits. At the same time, our lecturer will introduce the overall construction of PaddlePaddle Open Source Community and EDL, an open source project of PaddlePaddle Linux Foundation.

Although artificial intelligence has reached a climax again with the development of deep learning, the current artificial intelligence is essentially a non-robust pattern matching, which is easy to be attacked and has no deep intelligence (such as theory and experiment, induction and deduction). It is urgent to develop the next generation of artificial intelligence. Based on the analysis and comparison of human intelligence and artificial intelligence, we have come to the conclusion that the direction of the next generation of artificial intelligence is logical understanding and physical understanding, here come some technical clues.

Federated learning has received extensive attention in industry and academia because it breaks down the barriers of data silos and improves the efficiency of machine learning on the premise of compliance, and is gradually moving towards practical production and application. However, the complexity of federated learning in computing and networking exists objectively. With rich experience in cloud computing and virtualization, VMware CTO Office proposed the concept of cloud native federated learning to solve the practical problems of federated learning in production environment. This speech introduces KubeFATE, an open source project hosted by Linux Foundation, and how to use KubeFATE to deploy, operate and manage industrial-level federated learning system FATE from the perspective of practical application, and discusses how federated learning can work in complex IT environment.

OpenMMLab computer vision open source algorithm system is an influential artificial intelligence open source project in the era of deep learning. It has opened up more than 10 algorithm libraries, covering more than 15 research directions, including image classification, object detection, semantic segmentation, motion recognition, etc. It supports 150+ algorithms and provides 1200+ pre-training models. OpenMMLab has won a total of more than 35,000 star on GitHub, attracting more than 500 code contributors from home and abroad, widely praised by academia and industry, adopted by champion teams in many international academic competitions, and widely used in the actual business of many top Internet enterprises and small and medium-sized enterprises.

Open source is manifested as technological innovation, but behind it is the germination of new economics, new management, new sociology, new political science, new law and other philosophical and social science ideas in the digital era. What is the rationality of open source, which sprouted, blossomed and bore fruit in the traditional industrial society and was finally recognized by the mainstream society after all the hardships? What kind of social consciousness will be generated by the existence of open source society and what impact will it have on the world in the future? Open source needs to be explained from an economic perspective.