Yesterday, Google released Cloud AutoML, which does not need to write code. The fully-trained AI tool is regarded as Google's full force in the field of machine learning services (MLaaS) to catch up with other technology giants. Microsoft and Amazon are in the field of machine learning services. There are actions, how are the three PK battles, and what are the products? At present, MLaaS is still in its infancy, and what are the future prospects?
Google yesterday announced the launch of AutoML Vision, a major improvement in its Machine Learning-as-a-Service (MLaaS), which seeks to bridge the gap in competition with Microsoft over the past year or so.
Amazon AWS also announced its new MLaaS tools and services at the AWS Re:Invent conference last November, allowing AI application developers to build smart applications on the AWS cloud.
MLaaS is still in its infancy today, but it may become the company's leading AI platform, and companies are more willing to hand over development details to others through cloud rental AI services.
The following summarizes the strategies and tactics of the three giants of Google, Microsoft and Amazon to see who is the winner and who is the loser.
MLaaS: Commitment and problems
Machine learning is difficult, especially in the field of complex deep learning. Deep neural networks train millions of data samples and analyze them through NVIDIA's GPUs to extract and identify the characteristics and categories of the data.
This is the dawn of the "AI era," so companies and government agencies are of course scrambling to avoid missing the "next big event." To do this, they must decide which projects to invest in, hire scarce talent, buy large numbers of servers and GPUs, clean the data for supervised learning, and then build and optimize their own deep neural networks (DNNs).
Sounds hard? Then, MLaaS offers a simpler alternative: take shortcuts and use pre-trained neural networks to process images, video, voice and natural language data provided by major cloud service providers.
Since you can write a cloud-based application and access a pre-trained network through a simple API, why spend time and money training a neural network yourself?
Google, Microsoft and AWS: their respective strengths and strategies
Figure: Google's Cloud AutoML provides a dashboard that allows developers to easily measure the accuracy of the AI ​​model.
Google MLaaS
Strategy: Leverage Google's leading expertise in artificial intelligence and deep learning (with more than 7,000 AI projects within Google and more than 1 million AI users worldwide), providing state-of-the-art development tools and the highest performance hardware platforms for AI development. Google's platform is entirely about developers, because Google's users are very different from Microsoft.
Tactics: Make TensorFlow the king of AI hardware and software.
Apply AI to the development of AI. Google claims that its recently released Google Cloud AutoML can greatly simplify the complex tasks of the DNN development process.
Instead of enhancing the pre-trained APIs with additional custom data (such as those provided by Microsoft), Cloud AutoML builds a custom deep learning model starting with the customer's own data.
AutoML comes with a very cool dashboard that allows you to easily view the performance of your model as you develop and debug your model. Google even uses internal data tagging as a service offering—a manual process that people think will eventually be automated by AI.
Expand Google’s influence outside the data center. The impact of Google AI has expanded to include edge and consumer devices, as well as autonomous vehicles, covering all AI development projects on the Google Cloud platform.
Microsoft MLaaS
Strategy: Leverage Microsoft's vast enterprise and government installation base, and its broad portfolio of productivity tools and business process tools to become the default provider of enterprise ML technology.
Tactics:
A rich machine learning API is provided to handle various data types, as different companies or organizations have different data types depending on their business. Enable users to extend the training data of neural networks to products, people, vocabulary, etc. that contain their own organization (Microsoft is the first company to follow this path, Microsoft now offers 29 APIs - many of which support custom DNNs) Training data).
Provide the highest performance machine learning framework for customers who need to build their own deep neural networks, especially for natural language processing customers.
Use AI to enhance all of Microsoft's products—for Office 365, Dynamics, Windows, and, ultimately, to provide intelligent capabilities for each product.
Amazon AWS MLaaS
Strategy: Use AWS's ultra-large scale and rich toolset to provide the most cost-effective development and deployment platform for AI applications.
Tactics:
First, provide tools for AWS for Amazon's large online business development tools and platforms. Tools developed for Alexa and Amazon's own e-commerce are now available to developers for easy building of chat bots or voice-activated products or services.
Provide the world's best development tools, such as MXNet framework, Lex, Rekognition and SageMaker, to reduce development difficulty. These tools are very sticky, ensuring that AWS becomes the platform of deployment after the development process is complete. SageMaker is especially interesting and provides a fully managed platform for the lifecycle of the entire machine learning development.
Provide each developer with the most cost-effective cloud infrastructure, no matter which CPU, GPU or AI framework the developer chooses.
to sum up
First, regardless of how well these AI services are, companies need to recognize the limitations of MLaaS. The problem of course is the details.
If a pre-trained network as a service does not adequately contain the faces, vocabulary and objects that you want to identify? If you want to run an AI application on your own infrastructure, keeping all the valuable data locally, at least it seems safe?
In either case, MLaaS may not be the ideal entry barrier for the business. Microsoft and Google are trying to address these functional limitations of MLaaS, but Google's approach can produce more accurate results - AutoML actually lets users build a custom AI model instead of simply providing a customizable pre-processing layer .
By the way, I was surprised to find that AutoML is running on NVIDIA GPUs, not on Google TPU (also known as GTP). It is expected that this situation will change soon.
Nevertheless, I believe that Google's strength in artificial intelligence will help the company reach and possibly exceed Microsoft's current leadership in MLaaS technology, and Microsoft's strength in the enterprise software market will help it achieve its AI through its application portfolio. Invest.
Putting it all together, I think Microsoft will win the traditional enterprise market, and Google and Amazon will continue to compete for leadership in the new AI application development cloud computing. The Google Cloud platform will host applications developed on TensorFlow (and Keras), while AWS may serve other AI developers and application hosting markets.
How do tech giants crack giant monopoly
The Economist recently published a commentary saying that the monopoly of the three giants Google, Microsoft and Facebook is not good for consumers and industry competition itself.
Once upon a time, being the boss of a Western technology company was an enviable job. With the continuous development, technology giants such as Google, Microsoft and Facebook are considered to be too large, anti-competitive, addictive and undermining democracy. Regulators fined them, politicians denounced them, and former supporters warned that their power is causing harm.
This slamming of technology is mostly misleading. The inference of "big companies must be evil" is wrong. Apple has become the most valuable public company in the world for the simple reason that people want to buy Apple products. There is not a strong correlation between smartphones and sullenness.
However, big technology platforms, especially Facebook, Google and Amazon, have indeed raised concerns about fair competition. This is partly because they often benefit from legal exemptions. Unlike publishers, Facebook and Google are rarely responsible for what users do on their platforms.
For many years, most Amazon buyers in the United States have not paid sales taxes. The giants are not just competing in the market. They are gradually becoming the market itself, providing the infrastructure (or “platformâ€) for most digital economies.
Many of their services seem to be free, but users provide data for the platform, which becomes the way users pay for it. Although these giants are already very strong, according to stock market valuations, the number of investors is expected to double or even triple in the next decade.
Therefore, there is reason to worry that the technology giant will use its own strength to protect and expand its dominant position and harm consumers. Policymakers face tricky tasks: limiting these tech giants, but not overkill.
These platforms dominate because of the “network effectâ€. The number is interlocking: Amazon's more sellers will attract more buyers, which will attract more sellers. It is estimated that Amazon accounts for more than 40% of total online shopping in the United States.
Facebook has more than 2 billion monthly users and controls the media industry. The company is inseparable from Google, and Google handles more than 90% of web searches in some countries. Facebook and Google control two-thirds of US online advertising revenue.
The threshold for entering the technology industry is rising. Facebook not only has the world's largest personal database, but also has its largest "social chart": the list and the connections between them. Amazon has more pricing information than other companies.
Amazon's Alexa and Google Assistant voice assistants will better control people's experience on the Internet. Chinese technology companies have the ability to compete, but they cannot easily reach Western consumers.
If this trend goes with the flow, consumers will suffer because of the lack of vitality in the technology industry. Startups will invest less, most good ideas will be acquired by the giants, and profits will be eaten by the giants.
There have been some signs. The European Commission accused Google of using its mobile operating system Android to provide its own applications.
Facebook has been buying companies that will attract users one day: Instagram, WhatsApp and, more recently, tbh, an app that lets teens anonymously send praise to others. Although Amazon's competition is still growing, the industry, from grocery to television, can prove that Amazon can find competitors and squeeze them out of the market.
There are two remedies. The first is to make better use of existing competition laws. Antitrusters should carefully study mergers to gauge whether a transaction is likely to offset a potential long-term threat. Such a review may have prevented Facebook from acquiring Instagram and Google's acquisition of navigation software Waze.
In order to ensure that the platform does not favor its own products, an oversight team can be set up to review the opponent's complaints, a bit like the 2001 independent "technical committee" against Microsoft's antitrust case.
Second, antitrusters need to rethink how the technology market works. Personal data is actually the currency in which customers purchase services. Through data, tech giants receive valuable information about users' behaviors, friends and buying habits in exchange for their products.
Just as the United States developed complex rules on intellectual property in the 19th century, it needed a new set of laws to manage the ownership of data in order to protect individual rights.
In essence, this means giving people more control over their information. If the user is willing, the key data should be provided to other companies in real time, because now European banks need to process customer account information.
Regulators may force platform companies to provide anonymous bulk data to competitors in exchange for fees, a bit like a compulsory license for a patent. Such data sharing needs can be adjusted according to the size of the enterprise: the larger the platform, the more sharing is needed. These mechanisms will allow giants to privately store data and suppress competition into user sharing and innovation.
Fiber Optic Distribution Box,Fiber Optic Breakout Box,Fibre Optic Breakout Box,Fibre Break Out Box
Cixi Dani Plastic Products Co.,Ltd , https://www.danifiberoptic.com