< Maschinen das Lernen lehren - Forschungsnetzwerk "Lernende Systeme"
26.06.2014 By: Amanda Caracaş, ETH Zürich

Machine Learning - Pioneering Research Network

Shifting the paradigm: pioneering research network in learning systems established between MPI-IS and ETH Zürich - first summer school took place in June


It is 2050 and the world has become a personalized space where large and complex information, 'big data', plays a central role in everyday life. We might be paying for regular goods with data, commuting in self-driving cars, or providing vital information for medical research with our own biological profiles. Today, big data is projected to drive enterprise IT spending to $242 billion, according to Gartner, the world's leading IT research company. In academia, expectations towards big data are particularly high, for the potential of big data lies in systems that can learn from it: machine learning, a branch of artificial intelligence. In order to create and develop the foundations upon which such visionary ideas can flourish, ETH and the Max Planck Institute for Intelligent Systems have joined to form the world's leading research network: 'Learning Systems'. Solid foundations are essential where pioneering work takes place. It is upon those pillars that creativity, learning and exchange occurs and evokes new ideas in visionaries for the benefit of an entire society. 

Computer science: quo vadis? The era of big data and machine learning 

The zettabyte age hallmarks a mass of digital information approaching exponential magnitude. In this era where more data surges in a year than has been collected throughout the entire history of mankind, big data is no longer merely a byword for sheer volume. Rather, it is a signifier for opportunity. The key to unlock the potential of big data will be the ability to fathom its complexity and to draw useful information for societal benefit: from personalized medicine and data payment systems to intelligent search engines and driver assistance technologies, these innovations depend on a strong scientific foundation which has the ability to give rise to universal applications. For Prof. Joachim Buhmann, "advancements in the core technologies of computer science are no longer triggered by previous 'consumers' of data processing, the pioneers of early big data, such as CERN nuclear physicists or astronauts. Nowadays, complex models are being applied to countless fields, such as to biology or medicine where data is being measured quantitatively and/or predictively." Analyst company, Gartner, emphasizes: "IT is no longer just about the IT function. Instead, it has become the catalyst for the next phase of innovation in personal and competitive business ecosystems." In the vanguard of the information age, a scientific branch promises to take center stage: machine learning.
Machine learning thrives on data. Where data sets are large, diverse, and fast-changing, machine learning helps finding patterns and making predictions. As novel as Google's self-driving cars, IBM's 'Watson' on natural language understanding, Adobe's 'marketing cloud' for enhanced marketing analysis or Daimler-Benz' automotive combustion system design may seem, these technological advancements all draw on a familiar ability: the intelligence to learn and adapt. Where for decades, computer science primarily focused on how to program computers, now, researchers are creating synergies to explore how computers can learn from large data sets. Systems of the new age are autonomous, intelligent, can assist in decision-making and have the potential to pervade every area of human life. They even go so far as to outdo their impact on society, business and the industry by challenging the very traditions of their own methodological foundation. A revolutionary spirit, evident in both technological innovation and empirical inference, captures the present zeitgeist. 

ETH's Computer Science Department: reaching new interdisciplinary heights 

The desire to understand is a core human need. However, in this day and age where the amount of data has sky-rocketed, the complexity found in the mass of information surpasses the handling capabilities of a single human. Prof. Thomas Hofmann points out that "computers which can process large amounts of data have two advantages: firstly, it will be possible to reach a level of precision which is necessary for researchers to enter new fields which until now have not been accessible. Secondly, the aim behind the development of new theories is to serve society with personalized solutions. Science now has the potential to create the tools that will not only cater to the mass but, more importantly, to the individual." This is especially promising in 'personalized medicine', where scientists will be able to analyze a person's physiological situation with its complex biochemistry. It also applies to human-machine interfaces, where learning and adaptation play a large role. Visual interaction, autonomous navigation and e-commerce solutions are only a few of the examples where artificial intelligence and machine learning are avant-garde in revealing new possibilities and benefits for research, business, industry, society, and the individual alike. Prof. Buhmann adds that the research efforts at ETH have a strong focus on exploring the fundamental questions underlying machine learning to provide the foundation on which concrete visions can prosper.
The ETH Computer Science Department will contribute to machine learning by finding methods to extract relevant information, whereby statistical learning theory as well as information theory will play a large role in expanding scientist’s current thought on learning systems. Further, research groups in computer vision and computer graphics will focus on perception-action-cycles for autonomous systems: biological systems interact with the world by perceiving relevant aspects of the environment and their own state, processing these aspects, continuously deciding what actions to take based upon this information and adapting to the said environment by updating their priors. The insights won will be fruitful to the development of synthetic autonomous systems and robotics. Lastly, the department will focus on robust perception in complex environments. In nature, a staggering array of sensing systems has developed which have the ability to adapt to changing surroundings. Examining and learning from perception will advance genuine scene understanding and minimally scripted animation.
Expanding the horizon of methodologies in computer science calls for new talents as once again, knowledge and learning is advancing humankind from the established to the questionable, from the known to the frontier. It is inevitable that classic computer science studies will adapt to the need for more graduates with interdisciplinary skills. Prospective scientists will focus more intently on data handling, analytics and complexity theory. Prof. Buhmann remarks that the interest at ETH in machine learning is evident: "A few years back, we started a course with 30 participants. Now, we have 200. The interdisciplinary diversity which is brought about by big data and machine learning is a large opportunity - for graduate students and faculty alike." 

Two luminaries: ETH Zurich and the Max Planck Institute 

Today's global village is typified by communities that foster a close to real-time exchange of information. Transcending the boundaries of a single research institution pushes the perimeter of knowledge spheres towards an extension of consciousness - a mindset that is crucial towards sustaining scholarly excellence and becoming an academic leader of global influence. Rather than generating isolated research shelters, academic institutions build bridges between knowledge networks, nurturing and nourishing talent. In addition, universities and institutions are constantly being challenged to re-consider their obligation towards their immediate environment to provide valuable solutions of high societal, economic and industrial relevance. Prof. Andreas Krause remarks that in today's world, in which research topics are increasing in complexity, it is essential for institutions to recognize the value of their researchers and to form cross-institutional collaborations. Novel possibilities arise to push the boundaries of current research and a vibrant environment of knowledge exchange and transfer is cultivated. Such a platform is provided by the 'Research Network on Learning Systems' summer school for PhD students, taking place from June 16-20, 2014, at ETH Zurich.
ETH and the Max Planck Society are both archetypes of academic quality and value. They collaborate on strategic partnerships, embrace worldwide associations with outstanding scientists, and contribute to excellent technology transfer. Innovation requires a constant change of the so-called obvious. Therefore, these leading houses have united to become the world’s principal center for autonomous intelligent systems, 'Learning Systems'. The collaboration between four ETH departments (Computer Science, Information Technology and Electrical Engineering, Mechanical and Process Engineering, Mathematics) and the Max Planck Institute for Intelligent Systems, has set itself the goal of answering the core scientific challenge of our times in science and engineering: how to deal with perception, learning and adaptation in complex systems. Application domains range from medical informatics and system biology to robot perception and control, entertainment, education, social computing and smart materials. To foster research efforts, especially among graduates, workshops and conferences will take place and both institutions provide ample access to research facilities. 

MPI-IS Scientists who presented their research at the summer school in Zurich:

- Michel Besserve
- Michael Black
- Peter Gehler
- Manuel Gomez-Rodriguez
- Philipp Hennig
- Stefan Schaal
- Bernhard Schölkopf
- Metin Sitti