As the annual indicator conference of global cloud computing, Amazon re: invent 2020 conference has been going on for two weeks. One of the hot topics——Artificial intelligence (AI)andMachine learning (ML)It is also full of new ideas and frequent highlights. Amazon cloud technology (Amazon Web Swami sivasubramanian, vice president of artificial intelligence (services), personally delivered a 2-hour keynote speech, giving a panoramic description of Amazon Web services (Amazon cloud technology) In the field of machine learningRe: invent – reinvent。
Moreover, Andy Jassy, CEO of Amazon cloud technology, has spent quite a long time in his overall description of Amazon cloud technology, focusing on machine learning and its “reshaping” in various industries. Now, let’s quickly browse the key points and new ideas with you.
Strong Computing Foundation
As in the past, the new release of product technology, the launch of new functions and new features are still “coming as scheduled”. “Providing a strong foundation” is not only the first chapter of the speech, but also the “foundation” of everything else. Here, Amazon cloud technology provides a wide range of computing examples, including both general-purpose examples supporting the application of general training and reasoning, and training examples for high-performance requirements, such as p4d or reasoning examples, such as INF1, g4dn, etc.Here, the following new releases deserve attention.
One isAmazon EC2 instance based on Habana Gaudi AI processorIt is specially used for high-intensity machine learning training. The test shows that it is 40% higher than the comprehensive cost performance of Amazon EC2 machine learning training example based on GPU!
The second isAmazon training, a self-developed machine learning training chip of Amazon cloud technologyAnd will be used in the Amazon EC2 instance or provide underlying support for the Amazon sagemaker machine learning development platform.
Third,Faster distributed training and provided as part of Amazon sagemaker capabilities.Compared with the previous model, distributed training greatly improves its speed, such as mask RCNN training in the field of visual recognition,Tensorflow’s training time was reduced from 28 minutes to 6 minutes and 13 seconds, and pytorch’s training time was reduced from 27 minutes to 6 minutes and 45 seconds,These will play an important role in AI applications such as automatic driving.
If you want to knowMore heavy new releases, clickhereWatch the wonderful video, Amazon developer advocate, Pahud Hsieh Will take you backmachine learningandInfrastructureAnd the wonderful content of the keynote speech, a detailed interpretation of Re: invent’s heavy new release!
Empowering developers and data scientists
“Delivering machine learning to every developer” has always been Amazon cloud’s vision in the field of machine learning. Reduce the complexity of their operations in machine learning development and implement them through Amazon sagemaker, an integrated machine learning development platform.
In this regard, Dr. Swami elaborated in the chapter “creating the shortest path to success” and released a series of new products. For example, Amazon sagemaker data Wrangler for data preparation, Amazon sagemaker feature store for retaining various model features, Amazon sagemaker pipelines for workflow automation,… And so on. These can be regarded as developers throughAmazon SageMakerThe platform is more convenient and can be used to develop machine learning“Eighteen weapons”。
One of the most noteworthy isAmazon SageMaker ClarifyThe release of,Designed to deal with “data bias”(bias, academic translation) “.In popular understanding, it means that people inadvertently collect and process incomplete and objective data for various reasons in their work, resulting in “bias” in the result output.
For example, in video program recommendation, only after analyzing and processing the data of “drama” programs, it is considered as the overall of “entertainment” and recommended. “Data bias” is an important topic in the social application of artificial intelligence and one of the foundations to ensure the objective and fair application of technology. Now, Amazon cloud technology runs bias detection through the whole ml workflow, which not only saves time for the development of relevant applications, but also improves the overall quality of the model. The drift generated in the process of model aging can be marked out.
“Expansion” of machine learning
Machine learning is very advanced, but it is still in its early stage and has a limited audience.Find a wider range of “outside the circle” developers to join in, so that they can “get started easily” based on the existing knowledge system, which is where the long-term development of machine learning lies.Amazon cloud technology clearly sees this. Therefore, in the third chapter “extending machine learning to more builders”, Dr. Swami led the team to releaseAmazon Aurora ML、Amazon Athena ML、Amazon Redshift ML、Amazon Neptune ML、Amazon QuickSight QAnd other products and services, which representRelational database, SQL query of Amazon Simple Storage Service (Amazon S3), data warehouse, graph database, business intelligenceEqual tomachine learningGather in one place“re:Invent”。
The most important thing is that developers can operate with machine learning functions only by using well-known SQL statements.This is possible because Amazon cloud technology has carried out “system integration” in the background. withAmazon Aurora MLFor example, when users query customer information through SQL to try to find some negative feedback, Amazon Aurora ML will automatically schedule AI services such as Amazon comprehensive to return AI supported query results.This greatly reduces the threshold for the wider application of machine learning, and makes the general academic technology really begin to enter the “public vision”.
Solve real business problems
An interesting phenomenon is that Andy Jassy and Dr. swami, CEO of Amazon cloud technology, have set up a chapter such as “industry application” or “solving business problems” in the second half of their speech. Andy quotedClay Christensen * The purpose of customers looking for products and services is to complete their work… I’m looking for you not for machine learning, but for a specific job. If you finish the work by machine learning, it’s good, or other methods can actually do. But I finally look for you to finish the work. “
In view of this,The release of Amazon cloud technology has also greatly expanded machine learning to industry applications.for exampleAmazon connect, which is applied in the field of customer service center, fully integrates various AI / ml services,In addition, new functional services such as Amazon connect wisdom, Amazon connect customer profiles, real time contact lens for Amazon connect,… Are added, so that the agent personnel can realize the real-time response in the interaction with incoming customers through machine learning support for the knowledge base, intelligent search for answers to customers’ questions, machine reading customers’ calls and analyzing semantics, Second response to improve customer service satisfaction.
Another exampleIn the field of industrial manufacturing, Amazon cloud technology released the Amazon monitron Series suite solution,It forms a whole through sensors (detecting vibration and temperature), gateways, Amazon cloud technology cloud, mobile applications, etc., providing end-to-end solutions for AI based device maintenance. This is of practical value for occasions such as generators, large machine tools, heavy equipment and so on. In addition, there are supporting releases of Amazon lookout for equipment, Amazon lookout for vision and other related services.In terms of industrial intelligent detection, the release of Amazon panorama appliance,It sets a new “benchmark” for the integrated application of computer vision, IOT and machine learning. In addition, there are… Space constraints, there will be no more expansion.
Re: invent is indeed a “feast”, which systematically shows Amazon’s progress in cloud computing over the past year.What is new this year is the “advance” in the above application fields, from which we seem to see how to implement “industry 4.0”, a combinationArtificial intelligence, man-machine collaboration, process optimization, sustainable iterative improvementThe figure of industrial manufacturing is very different from the existing general cognition!
*Clay Christensen, author of disruptive innovation, is called “one of the most influential business theories in the past 50 years” by Forbes. Ranked third in the Global Thinker 50 rankings in 2017.
Amazon re: invent China website speech video is now online!
Amazon re: invent 2020 invites you to start together!
Dear developers, scan the QR code in the figure below or clickhereSign up for the meeting~ (PS: Chinese subtitles)