2019 Machine Learning: Tracking the path of AI development

2019 Machine Learning: Tracking the path of AI development

Https://mp.weixin.qq.com/s/HvAlEohfSEJMzRkH3zZtlw

The time has come to "guide" the "Smart assistant". Machine learning has become one of the key elements of the global digital transformation, and in the enterprise domain, the growth of machine learning use cases has also been significant over the past few years. The enterprise adoption rate for machine learning tools and solutions is expected to reach 65% by the end of the decade-and spending will reach $46 billion (according to IDC reports). On average, 55% of enterprise CIOs have seen machine learning as one of the core priorities for business acceleration. Here, we will focus on how machine learning will continue to evolve in the 2019.

Author |hussain Fakhruddin

compiling | Special knowledge

Finishing |yingying, Li Dahuan

New use cases for machine learning are coming up

Earlier this year, the U.S. Army announced that it would use custom machine learning software tools for predictive maintenance of combat vehicles. In other words, machine learning will be able to predict when and what type of service a vehicle may need. Another interesting machine learning use case is predicting stock market volatility based on previous stock earnings records. A recent study has shown that using machine learning to predict the stock market has more than 60% accuracy. In the area of medical health, machine learning models are used to estimate the probability of a person's death (in this case, the accuracy rate is far greater than 90%). Retail, marketing and sales as well as industrial/manufacturing scenarios also often occur in machine learning cases. "read" and "interpret" past data and predict the future -this is the nature of machine learning and technology will surely become more sophisticated.

Note: The concept of AI applications and machine learning tools is no longer confined to robotics. Instead, they have become a natural extension of business workflows and everyday applications.

Using ' Hardware optimized for machine learning ' will appear

2019 is likely to be specially prepared for silicon chips-with custom AI and machine learning capabilities-to become mainstream, at least for the enterprise. The AI optimization hardware market will continue to grow rapidly for the foreseeable future. A series of new, powerful processing devices will appear-we can also see high-end CPUs and GPUs. In summary, these tools and platforms will greatly enhance the usability of machine learning hardware.

Cloud computing combined with machine learning

By 2020, the global cloud computing market is growing at an annual rate of about 25%. The growing popularity of machine learning in enterprises is a key factor driving this surge. In order to successfully implement the "machine learning culture", companies must focus more on innovation than ever before-with special emphasis on improved cloud hosting and infrastructure parameters. Over time, more and more "AI-specific tools and systems" must be stored on the cloud-the latter requiring sufficient security and availability standards. Powerful, scalable cloud support will help organizations seamlessly move from machine learning to deep learning, delivering greater value to end users and improving their ROI data.

Note: Starting in 2019, general users will begin to understand more clearly how AI and machine learning processes work. Given the critical nature of artificial intelligence in areas where it exists, such as medical science, it is natural for people to know how technology can draw its conclusions/forecasts.

Continue to promote the capsule network

The advantage of neural networks is that they typically do not consider the relative direction or position of the selected object. Therefore, there may be an "information gap". And the capsule network is meant to be born. They are likely to replace many traditional neural networks in 2019 and beyond. In terms of performance, these capsule networks are more advantageous than traditional neural network systems-with more accurate pattern detection, and in the case of small amounts of data, the error probability is greatly reduced. More importantly-the capsule network does not need to repeat the training iterations to "understand" the changes.

Note: The Advanced Health care module based on machine learning algorithms is used to compare patient medical images and other medical images. The biopharmaceutical company AstraZeneca (AstraZeneca) plans to use robotics and machine learning extensively-to develop intelligent diagnostic systems in China.

The rise and rise of AI assistants

Siri and Google smart Assistants and Alexa have become part of our daily lives, and more importantly, each of the top "smart assistants" has become smarter each year (based on 5,000 general questions, Siri managed to answer about 31%, of which nearly 80% were the correct answer; In the same survey, the Google Smart Assistant answered more than 67% of the questions with an accuracy of less than 88%. As the machine learning range expands, AI Assistants are ready to go beyond the smart home. Starting next year, Hyundai and Kia will begin to offer built-in AI virtual helper systems in their new models. These assistants will be able to perform countless tasks-from remote home and car control functions (via voice) to destination recommendations (based on previous preferences) and navigation guides. In all areas of life, "smart assistants" with machine learning capabilities will make life easier than ever.

Note: Intelligent chat robots (with artificial intelligence) are also rapidly emerging. However, it is necessary to be vigilant-because the deviations in the training data set can cause serious damage to the user experience. Microsoft's ' Tay ' chat robot is a classic example of this failure.

Developers will focus on using machine learning to solve more "real problems"

When it comes to technologies such as artificial intelligence (multipurpose drones and auto-monitoring cameras and autonomous cars), it's easy to overdo it. However, it is important to realize-although all of these things can be true-but the steps of a mature data-driven ecosystem must be gradual and systematic. In the 2019, application developers and AI experts will focus on using machine learning to successfully address truly important needs (individuals and businesses)-rather than simply creating new deep learning tools prototypes. In other words, developers must understand that artificial intelligence and machine learning are more than just a few technical buzzwords-if implemented properly, their potential may be endless. There are many other technologies currently vying for attention (such as 4d printing), unless the development of AI solves the real problem, otherwise investors may start looking elsewhere. Separating "Ai Overhype" from "Artificial Intelligence" is critical and takes action based on the latter.

Note: In a recent study, 89% of CIOs are planning to implement machine learning tools and applications in their business.

The world of robots?

The role of intelligent robots in the workplace is increasing-and the improvement of machine learning is the main reason. In Japan, by 2025, artificial intelligence robots will provide three-fourths of the elderly care services-replacing human caregivers. Tianyuan Clothing-A Chinese T-shirt company-plans to use "sewing robots" at its Arkansas plant. In general, many labour-intensive tasks, especially repetitive activities that do not require too many professional skills, will be implemented by "intelligent robots" in the near future. In addition to making workflow smarter, increasing availability and reliability, and shortening time-to-market, machine learning-driven robots can significantly reduce operating costs (as well as outsourcing costs, if any). Increasing productivity should be a direct result of the full adoption of artificial intelligence in the workplace.

Note: Machine learning can also play an important role in precision agriculture. Smart poles for agriculture, with deep-rooted sensors and dedicated machine learning modules, can help farmers make smarter decisions.

Voice Technology stands out

It remains to be seen whether comscore predicts that 50% of the search activity will be supported by voice by 2020-but speech recognition (and the interaction based on it) has become an important fact that cannot be freed from the elements of machine learning. Unlike earlier speech technology, today's speech recognition ber is less than 5%--which is more than available. Interactive Voice Response (IVR) systems have become smarter than ever-because of iterative learning, voice-based machine learning systems can transcribe a variety of languages/accents. The trend for developers to launch mobile applications that support voice technology is also expected to gain further momentum in 2019. The Amazon Alexa and Google home assistants have understood our voice commands-they are paving the way for more of these platforms. Enter the market.

Note: Traditional, well-suited customer service executives are gradually being replaced by virtual roles. The latter provides a quicker response-and because the conversation is intelligent (virtual agents learn from previous conversations), personal touch is not lost.

The American and Chinese artificial intelligence market-war?

In the case of AI research and adoption, North America has traditionally been the front runner. However, the constraints are becoming weaker-the Chinese market is becoming a powerful force. In 2017, AI startups had a higher share of equity financing in China than their US counterparts (48% to 38%). China's AI start-up scenario is holistic (unlike the slight fragmentation of the North American market)-focusing on logistics, smart city projects, retail, healthcare, smart agriculture and other areas. In deep learning, China is clearly weakening it-releasing 6 times times more patients than the United States. According to the report, China hopes to compete with American AI in the 2020 and become an undisputed leader in machine learning technology within 10 years. It would be interesting to see how the US and China compete for global AI/machine learning hegemony in the coming years.

Note: Instead of relying on third-party APIs, developers are increasingly turning to making their own APIs for machine learning applications. There are many developer-friendly assembly kits and mobile SDKs to provide the necessary help.

More machine learning platforms (and a better platform?) ）

Platforms such as TENSORFLOW,H2O, Ai-one and torch have played an important role in how to deploy machine learning capabilities in different scenarios. In the coming year, we can expect a more powerful machine learning platform-with top-notch analysis, classification, and predictive capabilities. The capacity of these platforms is used in conjunction with other APIs, and big data will continue to improve. The continuous development of machine learning provides computers and mobile devices with the opportunity to "learn" faster and to better "interpret/analyze" data.

Note: AI/machine learning applications are also promoting automated decision management practices. Informatica and Uipath are good examples.

Radically changing the way people interact with technology

They may only be in a few places at the moment-but the ' no-cashier ' Amazon go ' store is revolutionizing the concept of shopping. In fact, by 2021, there were more than 2000 "Amazon go" stores in the United States alone. The way we work with smart things (especially) and technology, interact, and live (in general) is being shaped by the AI & Machine learning Revolution. Whether it's a business or a social or smart home-deep learning will disrupt our lives and ensure a total increase in efficiency. Through artificial intelligence, sci-fi movies and our imaginations seem to have become possible. The key here is the adaptability of the technology to different types of use cases. Machine learning is solving problems and providing value-and that's why it's getting more and more popular.

Note: The development of "killer robots" for war may be shocking. A recent report predicts that increasing investment in artificial intelligence in military applications is likely to lead to a nuclear war between 2040 and 2050.

NLP become more subtle

As a sub-domain of artificial intelligence, the importance of natural language processing (NLP) has increased significantly over the past few years. Natural language generation is primarily used to convert data into text, which is a key feature of many deep learning systems-and is used to write detailed market summaries or reports-NLP is very handy. The accuracy of natural language processing is also increasing, and automation systems are able to communicate ideas in a seamless way. Cambridge semantics and Attivio are some of the names of companies that provide NLP services.

Note: NLP modules typically need to analyze three aspects: syntax, semantics, and context. As more advances in machine learning and new areas of application are excavated, the demand for AI specialists, rather than technical generalists, will continue to grow. There are some grey areas-such as the prospect of mass unemployment and the possibility of intrusive surveillance-but, to be sure, 2019 will be an important year for machine learning. The age of Ai-as-a-service has arrived!

Original link:

http://teks.co.in/site/blog/machine-learning-in-2019-tracing-the-artificial-intelligence-growth-path/

-end-

2019 Machine Learning: Tracking the path of AI development