One of the biggest difficulties in talking about the use of machine learning is the lack of expertise. Data Scientists are desperately wanted everywhere. For this reason too, only large corporations can currently use ML.
However, machine learning can not realize its full potential through this limited use. The technology must be available to the mass. Therefore, the democratization of machine learning has the greatest importance for research and development. Barriers must be removed and the know-how transferred to all companies and institutions – regardless of their size. For this reason, the promotion of open technologies and at the same time a basic education is essential, so that the general public gets a minimum of feeling for the functioning and possibility of new technologies.
Only when the basic knowledge is available, the technology can be used in a wide range of applications that meet challenges in everyday life. However, the fact that comparatively little has yet been put into applications is not due to the technology. Rather, no concrete programs arise because employees are not yet thinking in this direction in order to develop their own systems for the company or the institution.
Each department should become an innovation lab
The instructions from above or the establishment of so-called innovation labs are not enough to develop new concepts. While this often gives rise to fresh ideas, infrequently there are machine learning applications that integrate with existing structures and offer important solutions to individual departments and the company. With the right understanding and fears of the new technology, each department can become a small innovation lab. The specialist departments create ML systems that can improve everyday life. Once machine learning and artificial intelligence are perceived as tools and opportunities, they can be used.
Where machine learning can be used
Incidentally, implementing a machine learning project by no means means going deep into programming. Modern tools make it possible to master even complex tasks with the help of clear surfaces and to train own models. For example, the Zoological Society of London captures thousands of animals worldwide with its camera traps. A model based on Google Cloud AutoML Vision recognizes these animals and classifies them. With this, the scholarly society can capture the species stock worldwide faster and even better protect the animals.
The simplified entry opens up completely new possibilities. Not infrequently, complete projects can be realized in half a day. The basic idea is to enrich even analogous areas with machine learning, making everyone an expert who can build and train their own systems. For example, a car mechanic might create an application that classifies paint damage by photo. Through a web application or app, customers could directly find out how much repairing could cost, and with reservations. This provides the end customer with a completely new service that can simultaneously improve the quality of the work.
Even if there is still a lot of potential in numerous companies, machine learning today has not only arrived in numerous industries, but has already become firmly established. Without a doubt, the manufacturing industry offers the most opportunities for machine learning in the German-speaking world.
However, many factories with production lines face challenges that are traditionally unknown to IT: many machines are not yet networked. Low bandwidth and poor mobile reception make it difficult to connect to applications in the cloud, where artificial systems have to be trained due to the high computational power. In addition, there is a high susceptibility to interference from radio links due to, for example, magnetic fields, as well as extremely high demands on safety and compliance. All these things make machine learning from the cloud difficult or seemingly impossible.
Machine learning offers many possibilities
Nevertheless, there are approaches to using machine learning even in these difficult conditions. Such systems can help optimize the entire production process or improve quality assurance. Just think of the visual evaluation of components. Modern technologies offer solutions by combining the best of two worlds. With a hybrid ML approach, the system is trained in the cloud with maximum performance and then realized locally with optimized microcomputers. The whole thing can then be used with extremely low latency, so that it can also be used in high-frequency production lines.
One example is the production of cookies and biscuits. Here, the quality of the cookies is permanently monitored by cameras. Has the dough been baked properly or are individual cookies too dark or too light? Was the marmalade applied circularly or was it spotted? High-resolution video streams provide machine learning systems with the data. In this way, cookies can be sorted out within a fraction of a second that do not meet the quality standard. Asynchronously, the newly collected data is used to permanently improve the model and train it in the cloud. This not only increases the quality of the cookies permanently, but quality assurance is accelerated and improved.
These examples show the possibilities of machine learning. Every industry can benefit from this and create their own systems as soon as the technology can be understood and implemented quickly. That’s why Machine Learning needs to be accessible to all. The democratization of technology is therefore the first most important step in realizing applications and solutions in a wide variety of areas. Only then can machine learning develop further and unfold the great potential for numerous companies.
Stefan Ebener (Customer Engineering Manager – Machine Learning, EMEA) is an EMEA-wide Machine Learning and AI Expert Team Manager for Google Cloud Customer Engineering. He also supports clients in the introduction, development and expansion of tailor-made solutions. Ebener is a freelance lecturer in business informatics and deals with the following topics: ML, AI and Big Data.