What are the shortcomings of machine learning

Navigation area

Machine learning has its origins in cybernetics and computer science, is a sub-area of ‚Äč‚Äčartificial intelligence (AI) and one of the most important technology building blocks in data science. Data science deals with techniques and theories of how hidden structures and implicit knowledge can be extracted from large amounts of data.

When it comes to machine learning, the Federal Ministry of Education and Research (BMBF) has focused on the constant promotion of software innovations. In many projects modern procedures were used and existing procedures improved. Funding from the BMBF over the past 20 years has proven to be a significant contribution to the fact that Germany is one of the leading scientific nations in the field of machine learning today. The scientific and technical progress in hardware and software has created steadily improved foundations for new and further developed methods of machine learning in recent years. New areas of application have developed on this basis.

New applications such as complex voice control, automated translations or autonomous driving show the technological and economic potential of modern machine learning processes. On the way to further, even more complex and highly relevant applications in international competition, however, the limits in the current state of science and thus open research fields of machine learning become apparent.

In order to shape top-level research in the field of machine learning in Germany for the future and to set new priorities, the BMBF, together with experts from science and business, identified relevant research fields in which concrete challenges for research are revealed. So there is z. For example, the challenge with autonomous systems is that the machine learning, which is important for environment recognition, has to be monitored by deterministic control and regulation systems that are understandable and explainable for liability cases. Further challenges are the lack of qualitatively acceptable data for machine learning in many practical areas as well as the rapid change in everyday environments, which prevents the collection of the necessary minimum amount of sample data. The evaluation of the data quality when compiling training data is another field of research. Even this small excerpt shows the need for research work on machine learning for everyday use in new fields of application.

The solution to the problem areas mentioned is of great strategic importance for the German economy, especially in the automotive, logistics, finance and medicine sectors. The BMBF will specifically support researchers and developers in subject areas such as deterministic behavior, transferability of knowledge, explainability of the solution, estimation of the error probability of a prediction, robustness of ML processes and increasing the efficiency of machine learning through various funding measures. Small and medium-sized enterprises (SMEs) in particular will benefit from the funding measures of the BMBF, since the training and further education of skilled workers as well as application-oriented pre-competitive development are to be promoted.