The prediction of the development trend of machine learning
Now, machine learning has really become an important part of the business vocabulary, and has brought wide and considerable potential development space for many enterprises. In 2017, after a series of avoidable setbacks, we expected that the machine learning ecosystem would eventually start to move in the right direction.
The prediction of the development trend of machine learning in 2017
The "age of analysis" is still in its infancy, which brings us a lot of ideas and promises that are worth looking forward to and excited about. In today's article, AtakanCetinsoy, vice president of BigML, will reveal the development trends in machine learning and related ecosystems in his eyes in 2017.
At the end of the year, technology experts will always focus on the new twelve months, and think about what the trend of technology plan will be in the next stage. In BigML, we try to analyze the future of the new year with the development and evolution of the machine learning technology in the 2016.
First of all, we need to emphasize that businesses need to blow away the hype surrounding the machine learning concept and explore effective ways to introduce them into their own business system. More specifically, enterprises need to choose platforms based on internal environment through rigorous decision making, and gradually establish smaller and easy to implement machine learning projects, so as to try to use their own data set. Over time, such incremental projects will bring positive feedback, and ultimately achieve decision-making automation, and even help agile machine learning team to completely change the normal operation of their industry.
In accordance with the usual practice, we first review the development process of machine learning technology in the practical application level:
Machine learning has formed an irreversible historical trend. We need to consider how to conduct cross department daily transaction processing and integrate our business with the overall market economy.
In the course of 36 years of development, many enterprises have been trying to digest, adopt and benefit from the progress of machine learning technology and related best practices. However, few companies can truly transform them into their own business advantages.
The emergence of a large number of so-called "new experts", they read a few books or in a few hall network course, began with a stately "change" in the world of cheap capital. At the same time, many top technology companies are recruiting talents who really understand machine learning skills as much as possible, hoping to store energy for the booming AI economy.
In addition, a considerable part of the start-up enterprise based on machine learning is the birth of mind "unicorn" ambition to embark on the journey, but must admit, universal, they think they can use the new machine learning algorithm to realize the magic of the low-cost, scalable solutions often just kind of wishful thinking.
In 2017, after a series of avoidable setbacks, we expected that the machine learning ecosystem would eventually start to move in the right direction.
Before we start to discuss specific predictions, we need to emphasize that 2016 is a very important year, because in the year, the five most valuable businesses in the world are all made by technology companies. The five companies have several common characteristics, including large-scale network effect, data centric corporate culture and new value-added services based on sophisticated analysis mode.
More importantly, these companies have been promoting their ideas and intentions and regard machine learning as an important fulcrum for their future evolution. With the accession of Unicorn enterprises such as Airbnb and the like, the leading position of technology industry in the world economy is likely to remain in the next few years, which will also be strongly promoted by the wave of large-scale digital transformation of the world economy.
However, this also raises a new problem may decide to trillions of dollars (for example: traditional enterprise holds the small non technology enterprise technology vendors and large amounts of data from large enterprises part dissolved and transformed) how to adapt and become part of the new value chain? How should they in living apart, to thrive in the new era?
At present, a considerable number of enterprises have to adhere to the rigid and experience guiding way of understanding the business intelligence system, continue to use the old traditional workstation class foundation, the running state of regression model statistical system, which means that it cannot capture the trend reflected in real life, not to mention the accurate prediction of the case complexity.
At the same time, these companies are faced with the plight of a large number of proprietary data that can not be fully utilized. According to the analysis released by the McKinsey Global Institute: "data driven world competition under the" report said, the modern analytical techniques mentioned in the report in 2011 has only achieved less than 30%, this is not all the technical scheme of the utility model over the past five years, the emergence of.
What is worse, data technology development trend among industries showing a serious imbalance (focusing on the United States, digital technology in the healthcare industry to adopt low to 10%, while the intelligent mobile phone field is as high as 60%), which means that the hitherto unknown analysis ability and the level of competition has emerged differentiation trend.
Although the actual situation is not up to the major suppliers and research enterprise publicity level (such as "cognitive computing" and "machine intelligence" or "smart machine" speculation concept), but the machine learning has really become an important part of business vocabulary, has wide development space and considerable potential for many enterprises. This great opportunity means that more traditions and start-ups will start their own journey of machine learning and exploration in 2017. Wise companies will try to draw lessons from failure cases and use new technology to expand their competitive advantages. However, considering the consistent folly and conservative attitude of mankind in the face of emerging things, we will discuss the following ten development trends in a more pessimistic way: prediction: machine learning will become an important way to achieve "big data".
The lessons of big data movement will be repeated again and again, and technical experts will also realize that only by combining various practical data solutions can they achieve their actual goals.
In general, the "big data" represents the data that can show the future, which is so simple. Gartner has recently eliminated "big data" items in its hype cycle report, which means it has formally entered the implementation phase. All these will emphasize highly the importance of analytical ability, especially machine learning plays an important role in guiding customers to use intelligent applications in data technology related projects. In addition, the previously criticized sample analysis scheme will become an important class of tools to help enterprises explore new predictive use cases in such application scenarios.
Forecast two: Vc firm will still actively provide funding for start-ups based on Algorithms
The Vc firm will continue to be in the state of groping and learning, and the whole learning process is slow and difficult. The venture will continue to provide funding for start-ups that have an academic settlement, ignoring misleading and even fantasies. For example, machine learning is a pronoun of deep learning, but totally ignores the great difference between machine learning algorithm, machine learning model, even model training and prediction results of trained models. The deep understanding of related disciplines will be a historic problem, and the importance of the overall investment industry is still not enough. However, it is worth noting that a small number of venture capital enterprises have begun to realize the huge development platform that the development of machine learning will bring.
Forecast three: machine learning talent will still be a hot and scarce resource
Media's advocacy and rendering of AI and machine learning technology will make the relevant technical talents continue to become the darling of the market, and the related investments will be concentrated in the hands of young scholars. But cruel reality tells us that most algorithms do not have wide applicability, and a large part of them only make a slight improvement on the original basis. As a direct result, most machine learning algorithms will be considered only as gimmicks and the reason for the crazy recruitment of technical talents. In some of the worst scenarios, buyers do not even have a clear analysis of technological development ideas, but just like following any trend of the times, they pay close attention to AI/ machine learning technology.
Forecast four: most of the machine learning related projects only stay at the PPT demonstration stage and do not bring the ideal results.
The traditional enterprise executives will actively hiring consultants to help them establish top-down analysis strategy and / or the development of complex big data technology components, but their feasibility for insight into the conclusion and the exact level of return on investment does not have the correct understanding. Part of the reason lies in the implementation of the correct data structure of data analysis technology and flexible computing infrastructure at present is not difficult to obtain, and after 36 years of continuous accumulation, now machine learning in cheap computing resources support is no longer a laboratory product too high to be reached.
Forecast five: deep learning paradigm in the field of business will be very few
Various well-known research achievements in deep learning, such as AlphaGo, will continue to attract media attention. However, the actual application plan represented by speech recognition and image cognition is the real development driving force, which will help the technology play a practical role in the machine learning scenario in the enterprise environment. It is difficult to explain, high level technology experts are scarce, highly dependent on large-scale training data set, and high computing resource allocation needs will restrict the development of deep learning in the 2017 years.
On the current situation, machine learning technology and polo is quite similar, it can bring rich and famous and exchange opportunities for you, also can let your enterprise instant forced bursting, but there are services, maintenance costs, equipment purchase costs and Equestrian Training expensive expensive club dues. Therefore, compared with the lack of significant research breakthrough and deep learning with unique advantages, enterprises usually can get faster and more realistic results by focusing on reinforcement learning and machine learning technology. Forecast six: the cause and planning based on uncertainty will push machine learning to a new height
Machine learning itself is only a small part of the AI. A considerable number of start-ups have begun to base on uncertainty to provide research and application for related reasons and planning exploration work, which will help us find new technology development space beyond mode cognition. Facebook's MarkZuckerberg, after damaging the AI/ machine learning research work for a year, has come up with its own personal intelligent assistant "Jarvis" -- its basic characteristics are similar to the fictitious intelligent housekeeper in Iron Man movie.
Forecast seven: Although the scope of machine learning continues to expand, human beings will still play a core role in decision-making.
Some companies will initially deploy machine learning programs that are faster and evidence-based decision-making, but human beings will still play a core role in decision-making. The application of intelligent early behind will focus on specific industries, but the difference of the regulatory framework and strict analysis ability of imbalance will give innovative management methods, the pressure of competition in the economic level, complexity of customer demand, high quality experience and some other value chain factors brought about conflicting guidance.
Although today's machine learning and artificial intelligence change the future, there is a lot of talk. But the sober technology leader is very clear that the real intelligent system will take a long time to really appear. At the same time, enterprises will gradually learn to trust their own models and their prediction conclusions, and realize that such programs can indeed bring more human performance to multiple tasks.
Forecast eight: Agility machine learning will quietly become the main force in AI marketing
A more realistic and agile machine learning approach will stealthily dominate the new year. The implementation team is willing to hand in hand and make full use of the rich enterprise data reserves, and can also completely bypass the "big data" related hype. They are more pragmatic and want to use the most targeted and applicable predictive means to solve the problem with small scale sampling data through a mature algorithm.
In this process, they will gradually build confidence in their own analytical capabilities, deploy the relevant solutions in the actual product, and add more practical use cases. Due to the restriction of more data access problems and deployment tool complexity, they can really use technology to enhance the core business data, experimental means and actively trying to risk and higher returns, consider predictive case as the approaches to achieve brand new revenue sources.
Forecast nine: MLaaS platform will become the "AI backbone" of machine learning in traditional enterprises.
The MLaaS platform will become the "AI backbone" in the field of accelerating the learning and practice of agility. Based on this, a new wave of application based on MLaaS infrastructure will further reduce the cost of business machine learning program, especially in the following ways: "democratization" in machine learning.
The cost is significantly reduced by eliminating the complexity of the supplier contract or the initial amount of input.
Provides a preconfigured framework that covers a large number of efficient algorithms.
It helps the end users to get rid of the complexity caused by infrastructure settings and management in an abstract way.
The easy and easy integration, workflow automation and deployment options are provided through RESTAPI and bundles.
Forecast ten: no matter whether there are enough data scientists or not, developers will continue to introduce more machine learning factors to their enterprises.
In the new year, developers will actively invest in machine learning camp, whether or not they have enough data scientists and other related talent reserves. Developers will quickly build and extend such applications based on MLaaS platform, and abstracting and stripping difficult details, such as cluster configuration and management, task queue, monitoring and distribution, etc. "Universal service" program will allow developers only through careful design and a good record of API can realize the machine learning technology, and no longer need to understand the LR (1) parser to compile and execute the Java code, or to realize the prediction of the case decision tree based on information gain or master Wilson rating mechanism.
At present, we are still in the early stage of development in the "analysis era". Therefore, we should keep an eye on the bright future instead of being defeated by some minor setbacks in the past. Although we in this paper put forward some rather pessimistic view, but this is purely in order to help the excitement of being friends to calm down and become dizzy with success, business success, to divide consciousness, mathematical mystery software and management best practices and implementation of scientific data has yet to cross the ability.