Amazon has launched AI to promote machine learning hosting services, four new tools, and AI hardware-Alibaba Cloud developer community.

Machine learning has developed amazing energy in many fields. Enterprises can efficiently lock users' needs and provide targeted services by obtaining effective user data, the pull-up effect of revenue profit is immediate.

However, for most enterprises that are in a hurry to get on the bus, the most difficult part of combining their own business with artificial intelligence technology is that, it is impossible to quickly find senior AI experts to analyze the business chain and build corresponding machine learning models to solve core problems and improve the efficiency of production or service links.

Amazon keenly captured this pain point. At today's innovation conference AWS Re:INVENT, the CEO of Amazon cloud service AWS Andy Jassy introduced this to more than 40,000 attendees. A complete set of hosting services that accelerate machine learning processes, SageMaker . And launch 4 big AI tools, video tracking tool Amazon Rekognition Video tool audio to text Amazon Transcribe emotional understanding Amazon Comprehend language Translation Amazon Translate.

SageMaker is an end-to-end machine learning service specially designed for enterprises and developers who want to add AI technology. This service allows data scientists, developers, and machine learning experts to quickly build, train, and host a certain scale of machine learning.


build Web applications with virtual learning environments from scratch for data mining, cleaning, and processing. Developers can run conventional instances or GPU-driven instances.

Model Training

build, train, and verify distributed models. You can directly use pre-installed supervised learning or unsupervised learning algorithms, or use the Docer Container Engine to train a model. This kind of training can process instances dozens of times, so the model building speed is extremely fast. The training data is read from S3 (full name Amazon Simple Storage Service), and the generated data is also put into S3. The data generated by the model is based on the model parameters, not the code calculated by the model. In this way, we can better use SageMaker to train models for other platforms, such as those IoT devices.

Model hosting

managed model services with HTTPs endpoints allow developers to perform real-time computations on models. These endpoints can relieve traffic pressure and can also perform A/B tests on multiple models at the same time. Similarly, developers can directly use the built-in SDK to build these endpoints, or use Docker images to set their own parameters.

"Boast, I think the most powerful part of SageMaker-end-to-end services is that these three parts can be used separately. , flexibly supplement and improve the existing machine learning workflow of the enterprise, "at the press conference, the CEO of AWS stressed the flexibility of SageMaker. "It not only provides ready-made tools, but also allows developers to build them themselves. No matter which option is selected, this service can use the most mainstream algorithm."

AWS CEO,Andy Jassy

the preset Jupyter Notebook has 10 built-in algorithms to solve many common machine learning problems. If you have special requirements, you can also build a machine learning algorithm framework, such as TensorFlow,MXNet, and Caffe.

Then, you can put the training data in AWS's Simple memory Service (Simple Storage Service, S3 for short). SageMaker processes all data, and then builds a data workflow, elastic block storage, and other elements. And then all treatment after them again split open.

In this way, developers can precisely fine-tune the performance of their models by optimizing the super parameters after baking.

"In the past, these tasks were manually operated, which was very tedious and time-consuming. Now AWS is much easier to worry about. You can test multiple parameters at the same time and then use machine learning to optimize this process." Jassy said.

Once the model is trained, developers can tell SageMaker how many virtual machines they want to use to test the model. In addition, you can also perform A/B tests on the SageMaker, allowing developers to intuitively see which parameter their model performs better after being changed.

SageMaker can solve the concerns of developers

collect and prepare data selection and optimization machine learning algorithm construction and management training environment training and adjustment model start to put the model into the production process promote the application of the model and manage and monitor at any time

now this service is free, but once the user exceeds a certain usage limit, the fee will be charged according to the usage frequency and region.

In addition to this AI cloud service, amazon AWS also launched four major new tools at the conference.

can identify specific people from multiple real-time monitoring streams and continuously track them. This function has now surpassed its competitors Google and Microsoft.

To match this algorithm, Amazon also launched an AI-driven DeepLens camera today. According to Amazon's previous performance of hardware sweeping the market, it can be predicted that DeepLens may be Amazon's next killer hardware.

Although Google also launched an AI-driven camera Clips two months ago, Google's camera serves more C- end consumers. Once something interesting happens, will automatically take photos and take photos. Amazon's DeepLens is for technical developers.

The DeepLens HD camera with about 250 dollars comes with pre-training models that will make it easier for developers to start recognizing text characters that appear in video streams. In addition, developers can also use AWS's new SageMaker AI services to train their own image recognition models and then run these models on cameras.

the human language in the audio file can be directly converted into text

nowadays, there are more and more audio contents on the network. How to identify and retrieve specific information from audio is a big problem.

This revolutionary engine launched by Amazon today can convert audio into text so that audio information can also be retrieved.

Q: What scenarios can I use to convert hot audio to text?

In many places, for example, you can get real-time subtitles in Japanese and Korean dramas in the future, and you can directly watch the cooked meat. You don't have to work hard to translate the subtitle group. Or enterprises that want to improve the service quality of the customer service center can no longer spend a lot of time listening to the telephone recording files one by one. It is much more efficient to directly read the text version.

However, currently, Amazon Transcribe only supports English and Spanish. However, Amazon officials said that a new version will be released in the next few weeks and more languages will be supported.

can recognize the positive or negative emotions behind the text from the words, context and character description. Currently, only English and Spanish are supported.

The first four functions are recognition language, noun classification, emotion analysis and key phrase extraction. These functions are developed for social interaction functions, with a response time of 100 milliseconds.

Amazon Comprehend requires constant training to provide better natural language processing services. Amazon's team of engineers and data scientists are making unremitting efforts to expand and refine the training data, hoping that everyone will use it more accurately in the future.

After acquiring Safafa's technology two years ago, Amazon finally launched its own language translation service. However, this service lags behind Google Microsoft for several years.

The technology is based on the language matching model represented in the neural network.

The model consists of two parts: encoding and decoding. The encoded part reads sentences from the language to be translated and creates an expression in the target language to match the meaning of the specified text. After creating a new expression, give it to the decoding part of the model to see if the generated expression conforms to the expression habits in the target language corpus and whether there is semantic deviation.

At the same time, in order to translate as accurately and concisely as possible, there is also a mechanism called Attention mechanism in this model. Pay attention to every word in the translated language text at any time, and combine the context to judge which words are to be translated into the target language and which can be thrown away.

Amazon hopes that this translation tool can be combined with other AWS services, such as text-to-speech Polly program, Elasticsearch tool for multi-language search, Lex chat tool, and content localization service provided through Amazon Lambda.

As CNBC reported earlier, the new service is likely to be based on Amazon's technology of purchasing Safafa two years ago. Today's announcement confirms these early reports and introduces AWS to translation services provided by Microsoft and Google.

According to Canalys, AWS led the cloud infrastructure service market by 31.8% in the third quarter. In this quarter, AWS brought Amazon $4.58 billion in revenue and more than $1 billion in operating revenue.

- finished -

Selected, One-Stop Store for Enterprise Applications
Support various scenarios to meet companies' needs at different stages of development

Start Building Today with a Free Trial to 50+ Products

Learn and experience the power of Alibaba Cloud.

Sign Up Now