SIEMENS

The amount of data produced worldwide is skyrocketing. According to market research company International Data Corporation (IDC), the digital universe — in other words, all digitally stored data worldwide — surpassed one zettabyte (10²¹ bytes, ZB) in 2010 for the first time. And IDC expects this figure to rise to 35 ZB by 2020 (see Pictures of the Future, Spring 2011, Zettabyte Goldmine). That is equivalent to the data contained in two piles of DVDs stretching from the Earth to the moon. Among the fastest-growing data categories are large data collections known as metadata, i.e. books and databases, as well as unstructured data such as arbitrary texts and graphics with an undefined structure. About one third of the digital universe currently consists of high-quality information — in other words, data and content subject to security, compliance, and storage regulations. IDC estimates that this kind of information will account for almost half of all data by 2020.

This growing mass of increasingly complex data must be efficiently processed. However, this is not possible without computers that help sort, analyze, and compress data, as well as preparing it for use by humans. Learning systems are particularly helpful in this regard, since they can learn from examples, recognize patterns in data, and use this information to predict future developments. The applications of machine learning are extremely diverse, ranging from market analyses and anticipatory maintenance in industrial applications to diagnostic methods for medical systems. In many cases, the focus is on technologies for voice, text, and image pattern recognition.

Voice recognition systems are used to operate vehicles, for example, as well as for automatic telephone switching, the management of building and office technology, industrial quality assurance, and medical diagnoses. Market researchers at Datamonitor expect high growth rates here in some fields. For example, they predict that the market for advanced mobile voice recognition systems in handsets will triple from $32.7 million in 2009 to around $100 million in 2015. According to these experts, the market for mobile voice recognition in automobiles will increase from $64.3 million to $208.2 million during the same period.

Voice recognition systems as such are nothing new. According to a report released by the market research company Gartner in 2011, voice recognition technologies were already part of the “hype cycle” of technologically relevant trends in 1995. The systems are still not fully effective, however, primarily because the recognition of colloquial speech is one of the biggest challenges that a computer can face. The main reason for this is that a computer needs to have extensive knowledge of everyday life in order to really understand what someone is saying.

Learning systems can also be used to analyze images and videos. Such systems are especially beneficial in industrial image processing. As a result, the European Machine Vision Association (EMVA) expects this market to grow by 20 percent in Europe in 2011, following an increase of 11 percent in 2010. Although inspections and quality assurance remain the most common areas of application for industrial image processing systems, new technologies are also being introduced — for example, in robotic 3D vision systems. These technologies range from video systems for automobiles to security solutions. Pattern recognition is, meanwhile, becoming more and more important in medical engineering (see Pictures of the Future, Spring 2011, “Hybrid Insights” p. 70). Business consultancy firm Frost & Sullivan points out that doctors are increasingly relying on learning software to filter out and process the key information produced by advanced digital imaging procedures, such as computer and magnetic resonance tomography and ultrasound systems. The software is used, for example, in mammography procedures as well as for the diagnosis of lung, pancreatic, and intestinal cancer.