Drag data into the graphical interface, AI will help you predict your business, and the new features of Amazon Cloud Technology are bright

Can you do machine learning without writing code? !

This is a new tool announced by Amazon Cloud Technology at the recent Amazon Cloud Technology re:Invent conference to implement codeless visual machine learning.

The Amazon Cloud Technology re:Invent conference can be described as the "Spring Festival Gala" of the cloud computing industry, a weather vane-level event in the IT technology field, and even their own Seattle headquarters building has also been renamed the conference. The brand influence is evident.

In the past 10 years, countless cloud computing and even AI industry benchmark products have been born at this event, such as Amazon Mechanical Turk, Amazon Rekognition, Amazon SageMaker and so on.

This year, Amazon Cloud Technology re:Invent 2021 also has a huge amount of information. The most interesting thing is that Amazon SageMaker, the machine learning platform service of Amazon Cloud Technology, has ushered in a "big explosion".

It is not only the “dish” of developing machine learning models without code. On the entire AI table, looking vertically, 12 products of Amazon Cloud Technology have covered the entire AI industry chain——

From a horizontal point of view, from the entry-level free computing pool for individual developers to the AI model optimization tools required by professionals from major manufacturers, the corresponding releases are also available. Even considering the rapid development of AI for users in the Chinese market, Amazon Cloud Technology also provides a conference record with Chinese subtitles at station B.

In the live speech, the CEO also specifically emphasized: "Amazon Cloud Technology will provide the most extensive and complete full-stack machine learning services."

may as well follow us to review the highlights of the whole process, and have a comprehensive understanding of the AI product context of Amazon Cloud Technology.

Machine learning without code

Let's first explore the code-free machine learning prediction service mentioned at the beginning to see if it can really be used by people who don't understand the code.

According to the official introduction, this product is called Amazon SageMaker Canvas , which is aimed at groups with zero machine learning experience. Among them, some may be business analysts, and some may be engaged in human resources, finance, or marketing.

It is foreseeable that most of the above groups have no machine learning experience or even knowledge of code, but they definitely have the need to use data to measure current strategies and predict market trends.

Amazon SageMaker Canvas is to visualize the many steps of a machine learning model as an interactive UI, aiming to solve their business problems, known as: Without writing a line of code, quickly generate a machine learning prediction model.

In order to prove its effectiveness, the AI/ML department of Amazon Cloud Technology shared a case.

Among them, the department’s product marketing experience ideally evaluates current marketing activities through Amazon SageMaker Canvas to determine whether it has sufficient influence and effectiveness.

Just open Amazon SageMaker Canvas and upload the data. During this process, the platform can also automatically correct errors in the uploaded data, such as adding missing values or deleting duplicate rows and columns. Its technology is not unexpected, and it also comes from its own AI/ML.

Next, specify the target predicted by the model, and then click "Quick Generate", the required model can be trained.

From the results, the rendering effect is indeed a visual chart, and the accuracy of the model is 93%.

After the model is generated, it can be shared with partners such as data scientists to help business personnel to further check or optimize these models.

After reading the official case, the visual interface does have two brushes——

So how does the partner experience?

At present, the BMW Group has put Amazon Cloud Technology AI/ML technology into more than 600 applications in the actual business process, covering multiple scenarios from the production line to the sales end. In addition, BMW also has 15 million connected cars involved, generating more than The millions of kilometers data are all analyzed and predicted by Amazon SageMaker Canvas.

Siemens Energy is also one of those who started eating crabs. They use Amazon SageMaker Canvas as a supplement to their own machine learning toolkit. A data science team leader in the application department said: Canvas allows us to share collaboration with the data science team, which helps to produce more machine learning models and ensure models Comply with quality standards and specifications.

There are also many little-known giants that are also Canvas experiencers. For example, INVISTA, a subsidiary of the Koch Group, the world's largest private company, has also used Amazon SageMaker Canvas to assist with data science issues in business processes.

reading the results of 161b84804a124e multi-party evaluation and intuitive display, it can be roughly judged that this Amazon SageMaker Canvas is indeed worth looking forward to. After all, the law that graphical interface releases productivity and creates value compared to code has been proven repeatedly in the past.

Free online AI laboratory

As mentioned in the previous article, at the annual blockbuster conference, Amazon Cloud Technology put down its arrogance: to provide the most extensive and complete full-stack machine learning service, since it is the "most extensive and most complete", only the release of Amazon SageMaker Canvas not enough--

For Canton University research institutions and AI enthusiasts, cutting-edge technology giants also need to live up to their own slogans.

summed up, three words, lower the threshold.

The most intuitive, provides computing resources.

In recent years, high hardware prices and complex software configurations have always hindered beginners from getting started with AI, and are also a huge obstacle that restricts the development of the industry and makes more people aware of it.

Amazon Cloud Technology releases the function Amazon Sagemaker Studio Lab provides a large group of "wool" that can be smashed. No additional environment configuration, no need to register an account, and you can log in to the online laboratory directly by email.

In this environment, anyone who creates a project can directly have 12 hours of CPU computing time, 4 hours of GPU computing time, and 15GB of storage space:

Looking at the entire industry, such a configuration is indeed in place.

You know, when using Pandas or XGBoost for data preprocessing for classic ML algorithm training, 12 hours of CPU time is basically enough. For deep learning training, you can also choose the GPU backend to get 4 hours of computing time, which is enough to train or fine-tune the model on a smaller data set.

In other words, for the beginner stage AI model, it can basically be trained for free with the above resources.

At the same time, the most popular machine learning tools, frameworks and libraries are also pre-packaged and provided to registrants. They can customize the Conda environment and install the open source JupyterLab and Jupyter Server extensions. The above-mentioned experimental environment is tightly integrated with GitHub, so that the created project can be easily copied and saved.

In addition to free "online labs" and computing resources, another part of "wool" is more intuitive-scholarships.

This time, Amazon Cloud Technology has provided a total of 10 million dollars to launch a 161b84804a13e3 Amazon Cloud Technology AI&ML Scholarship Program , which aims to help high school and college students over 16 years old and pave the way to machine learning-related careers.

In addition, Amazon DeepRacer, the 1:18 ratio self-driving race car of Amazon Cloud Technology, is also for self-driving and machine learning enthusiasts, providing a more interesting and lower barrier to help them get started with machine learning and train Own reinforcement learning model.

Amazon DeepRacer is driven by reinforcement learning and can deploy algorithms in a 3D racing simulator in the cloud. It can also experience the excitement of racing in the real world through a physical car.

Of course, those who perform well can also pass through the scholarship program.

Not only does it shine, but Amazon Cloud Technology also pulls in Intel, and Udacity launched a joint event to distribute 2500 scholarships to disadvantaged social groups such as economic difficulties and disabilities over 16 years old.

In addition to obtaining financial support, these disadvantaged groups can also receive guidance and help from Udacity mentors, Amazon Cloud Technology and Intel technology experts for up to a year.

Machine Learning "Industrialization" Reshaping

No matter the release of zero-code machine learning or the inclusiveness for a wider audience, there is still technology behind it. After all, functional development requires deep scene understanding and technological accumulation, and the test of the word "inclusiveness" is the cost reduction level of technology companies.

Compared with the above two, Amazon Cloud Technology re:Invent2021 released the new features of Amazon SageMaker for professional practitioners, which more intuitively demonstrates the technical level of Amazon Cloud technology, from which it can be seen that the technology giants plan for the future of AI/ML.

For the majority of MLers, a complete machine learning process includes data preparation, data labeling, training, inference, and deployment. The final model inference effect depends not only on the personal level of the developer, but also on external factors such as architecture, computing power, and data.

The reason why Amazon Cloud Technology does this is to reduce the impact on the individual level, in their words: let AI/ML move from handcraft to industrialization.

Specifically, in order to solve the problem in a package, Amazon SageMaker gives a set of combo punches, covering the whole process of machine learning:

In the data preparation , data engineers often need to leave the current development environment and manually configure a cluster that meets the running model or analysis requirements.

To this end, Amazon SageMaker Studio is integrated with Amazon EMR, and SparkUI can be used directly from Amazon SageMaker Studio Notebook to monitor and debug Spark jobs running on Amazon ECR clusters.

In view of the fact that no matter whether you perform data preprocessing, development or model deployment, you do not need to leave this environment, the above actions are undoubtedly a step forward to an ideal fully integrated development environment.

data annotation stage is also saying goodbye to labor-intensive, to avoid artificial succumbing to artificial intelligence:

This task used to require manual labeling by humans or processed by data labeling programs, but now, after the original data and requirements are given, Amazon SageMaker Ground Truth Plus will combine pre-labeling assisted by machine learning to assist human experts in labeling.

This method can reduce the error rate and at the same time reduce the cost of labeling by 40%, so as to detect errors more efficiently and avoid the appearance of low-quality labels.

training phase is more critical.

Stronger than the industry's classic deep learning model BERT, a complex neural network with billions of parameters requires thousands of hours of GPU training. Even if the parameters are adjusted and optimized, it still takes a few days to train.

But now, Amazon SageMaker Training Compiler, a machine learning model optimization compiler provided by Amazon Cloud Technology, can improve GPU instance training speed without adding too much code.

With the help of this compiler, many classic deep learning models including BERT-base-cased, BERT-base-uncased, and distilBERT-base-uncased can directly increase the training speed by 50%.

Add two lines of code to use Amazon SageMaker to train the compiler

Finally, there is an inference stage . Amazon Cloud Technology came up with the previously famous "serverless" concept and provided a set of Serverless Inference with serverless inference functions.

This function is able to allocate resources to the cloud and enjoy a flexible resource space service for the situation where the amount of data calculation is highly volatile. Let programmers focus on high-level languages instead of low-level hardware, allowing professionals to focus on what they are good at.

Considering that in reality, many customers have special needs, but it is difficult to judge how much computing resources are appropriate. Another function, Amazon SageMaker Inference Recommender, provides configuration and actual operating parameter recommendations in the inference phase to find the best balance between cost and speed. point.

From the data preparation to the inference stage, the above-mentioned various process product functions are released as full machine learning cycle services, rather than a single point patchwork. The purpose is to help enterprises realize the large-scale application of machine learning. AI/ML industrialized scale application process.

So what is the effect of this combination boxing?

In the visible case, Vanguard, one of the largest fund management companies in the United States, has reduced its deployment time by 96%. Pharmaceutical giant AstraZeneca can complete the deployment of its machine learning environment within 5 minutes. The financial management company NerdWallet has reduced its cost by 75 under the premise of increased training requirements. %.

In addition, more diverse landing scenarios can also see Amazon Cloud Technology's in-depth exploration of AI/ML.

For example, DevOps Guru for RDS can be used to help developers detect, diagnose, and solve performance and operational issues in Amazon Aurora.

For example, CodeGuru Reviewer can identify passwords, API keys, SSH keys, and access tokens in the source code to improve the efficiency of code review and help the traditional software industry to improve its efficiency.

What's interesting is that during the re:Invent 2021 conference of Amazon Cloud Technology, CTO Werner Vogels also published a blog in a hurry, which exposed the high expectations of the technical man for the AI/ML industry:

Software development will begin to shift from labor intensive, and software development supported by artificial intelligence will occupy a dominant position.

Finally, in terms of hardware, Amazon Cloud Technology also released a self-developed chip, and launched three models in one go.

Among them, the CPU chip Graviton3 is based on machine learning.

There is also a machine learning customized training chip, Trainium, which supports Trn1 instances, which can provide users with more cost-effective and faster speeds for training deep learning models in the cloud.

Whether it is opening up the AI/ML industrialized scale application process or the release of hardware self-developed chips, looking at it at a more macro level-

The above announcement shows the visible extension of Amazon Cloud Technology in the AI/ML business.

Amazon Cloud Technology is expanding the boundaries of AI

According to IDC data, in the seven years from 2013 to 2020, the scale of global AI/ML annual expenditures has rapidly expanded from zero to approximately US$50 billion. This growth rate is almost twice that of Amazon Cloud Technology’s old bank cloud computing.

It is precisely because of this trend that Amazon Cloud Technology's multiple attacks seem inevitable.

From code-free machine learning and serverless applications to AI/ML, to the continuous upgrade of underlying computing power, and even many inclusive programs... The dizzying releases all show that Amazon Cloud Technology is redefining a new frontier of machine learning.

Although the results of the above release have not yet landed, the value displayed by the general public may not be visible to the naked eye for a while, but from another perspective, the so-called pursuit of long-term value and the so-called importance of infrastructure layout are obviously written in the DNA of Amazon Cloud Technology. ?

Recall that in 1997, Bezos issued the well-known "Letter to Shareholders."

At that time, the Internet bubble of the first generation was accumulating, and "quick money" was still a new concept to many people. At that time, Amazon proposed that customers, sales, and brand growth are all for long-term value services, and also long-term value. Bezos emphasized that " Continuous investment in systems and other infrastructure.

Since then, Amazon Web Services has operated independently, turning cloud computing from a "concept" into a real industry, and Amazon Redshift and Amazon Lambda have brought out cloud-native data warehouses and serverless development...

Everything seems to have written the prologue early.

Now that Amazon Cloud Technology continues to bet on the AI/ML field with the same mentality, it is not surprising in itself.

This is not only the responsibility of being a technology giant, but it also makes people look forward to it. It is echoing the Slogan of this year's Amazon Cloud Technology re:Invent 2021: Leading the wind and reshaping the future.

We can already see that the population coverage in the AI/ML field is expanding, its industry scenarios are expanding, and its technology continues to be explored accordingly. This process is still being continuously promoted by industry technology pathfinders.

future, how big is the territory of AI/ML? Amazon Cloud Technology is drawing a new outline little by little.

Portal to the full content of the conference (long press to scan the code):

​