[Meetup Preview] OpenMLDB+OneFlow: Link feature engineering to model training to accelerate machine learning model development

On July 31, 2022 (Sunday), from 14:00-16:30 pm, the fifth meetup of OpenMLDB, an open source machine learning database, will be broadcast live online.

Event Background

OpenMLDB, an open source learning database that provides full-stack solutions for production-level real-time data and feature development, invites static compilation and streaming parallel deep learning framework OneFlow to collaborate to bring the fifth session of OpenMLDB Meetup.

This online sharing will lead you to deeply understand the iteratively upgraded OpenMLDB and OneFlow, analyze the architectural ideas and hard-core technologies behind the products, and demonstrate how to easily calculate features through OpenMLDB, and combine OneFlow to smoothly train models to accelerate machine learningModel development, help AI low-threshold and low-cost landing!

Brief introduction

OpenMLDB PMC core member Lu Mian will start from the low-cost, high-performance open source solution of online and offline consistency feature platform, and introduce the latest version of OpenMLDB and its performance improvement, cost reduction and flexibility increase.characteristic.

OneFlow PMC core member Chengcheng will focus on OneFlow - Making Large-scale Distributed Deep Learning More Convenient, and introduce to the audience that ease of use and completeness are further improved, model migration is more convenient and fast, and large model support is available.More efficient OneFlow v0.8.0 and other highly available and scalable solutions and components.

Huang Wei, OpenMLDB system architect, will demonstrate how to calculate features through OpenMLDB and how to use OneFlow to load and train feature data, and use practical exercises to show how to combine OpenMLDB and OneFlow to easily implement feature calculation and model training.

Deng Long, Platform Architect of OpenMLDB, will deeply analyze the hard-core technology behind the architecture design of OpenMLDB, and guide you to understand the internal implementation of OpenMLDB's millisecond-level real-time online feature calculation engine.

See the poster for the specific schedule. The live broadcast information will be synchronized in the OpenMLDB technical exchange group. Friends who have not joined the group are welcome to join the group to watch~

[External link image transfer failed, the source site may have anti-leech mechanism, it is recommended to save the image and upload it directly (img-qV8nqG1f-1658928990164) (https://p9-juejin.byteimg.com/tos-cn-i-k3u1fbpfcp/b598c277c026405d98a860305ce334d8~tplv-k3u1fbpfcp-watermark.image?)]

Share a sneak peek

OpenMLDB: Consistent production-grade feature platform online and offline

【Speech Outline】

1. Data and feature challenges for AI engineering implementation
2. OpenMLDB: Consistent online and offline production-level feature computing platform
3. Introduction to new features in v0.5.0: Improved performance, reduced cost, increased flexibility

【audience benefit】

1. Understand the pain points of data and features encountered by enterprises in the process of implementing artificial intelligence engineering
2. Learn about the low-cost, high-performance online and offline consistent feature platform open source solution: OpenMLDB
3. Understand OpenMLDB's online and offline consistent design architecture concept and enterprise-level product features
4. Learn about OpenMLDB v0.5.0 features, improved performance, reduced cost, and increased flexibility

OneFlow - Making Large-Scale Distributed Deep Learning Easier

【Speech Outline】

1. Interpretation of the new version of OneFlow v0.8.0
2. Global Tensor: An easy-to-use solution for distributed execution brought to the community by OneFlow
3. Graph: an efficient and fast dynamic and static conversion solution, providing an easy-to-use advanced distributed parallel optimization configuration
4. LiBai: An efficient and scalable large-scale distributed pre-training code base based on OneFlow
5. OneEmbedding: An efficient and flexible extension component designed for large-scale recommender systems

【audience benefit】

1. Learn about the easy-to-use solutions for distributed execution provided by OneFlow
2. Understand high-level optimization techniques and the nature of distributed parallelism in large-scale distributed parallel training
3. Learn about the features of LiBai's large-scale pre-trained model library and its advantages over other solutions in the industry
4. Understand the features and advantages of OneEmbedding in solving large-scale recommender system problems

OpenMLDB+OneFlow, teach you how to quickly link feature engineering to model training

【Speech Outline】

1. Demonstrate using OpenMLDB to compute features and OneFlow to load feature data for training

【audience benefit】

1. Learn how to compute features with OpenMLDB
2. Learn how to use OneFlow to load feature data and train
3. Learn how OpenMLDB and OneFlow work together

Demystifying the OpenMLDB millisecond real-time online feature calculation engine

【Speech Outline】

1. OpenMLDB Online Architecture
2. Design and implementation of storage engine
3. The principle of building a highly available database

【audience benefit】

1. Understand the overall architecture design of OpenMLDB
2. Learn about the implementation path of the millisecond-level real-time online feature calculation engine