Teaching Machines How to Learn to Improve Business and Life

, by Riccardo Zecchina - Vodafone chair in data science and machine learning
Big data sets and algorithmic analytics have accelerated scientific and technological progress and already improved our lives. But the challenge of machine learning has only begun and younger European generations must equip themselves to not lag behind in a game currently led by the United States and China

The proliferation of data and the invention of powerful algorithmic tools for their analysis are powerfully accelerating progress in many areas of science and technology. Computers are now able to recognize objects within complex scenes, process voice expressions and answer human questions, identify relevant aspects of huge data sets in various domains, and compete in complex games that require forms of intuition to deploy sophisticated strategies. In many applications, machine learning is reaching a level of performance comparable to, if not better than, human counterparts. More importantly, machine learning can do so on a large scale. Life sciences and social sciences, both characterized by complex sets of data, are at the cusp of this great revolution. Machine learning will have a huge impact on business and personalized medicine, as demonstrated by revolutionary applications already in use, such as human-machine interaction in natural language, and the use of machine learning tools to predict outcomes of novel gene-editing techniques (the so-called CRISPR revolution in biotech).

In spite of recurrent media exaggeration over the last few decades, all these studies and applications crucially depend on data and were still unattainable even as recently as ten years ago. Actual progress has been triggered by the combined development of new technologies for the production and mining of data, along with powerful computer platforms and new machine learning algorithms. Currently deep neural networks are the main tools of machine learning, although not the only ones. These artificial systems are a special case of a wide array of machine learning techniques that mimic information and communication processing in the visual nervous system. Their goal is to extract relevant information from huge amounts of data (i.e. to learn so-called data representation) and machine learning has already been applied to a multitude of fields including (but not limited to) computer vision, speech recognition, processing natural language, audio recognition, social networks, machine translation, bioinformatics, drug design, and data analysis in general.

Most of these questions have been studied for decades, but only recently has machine learning produced concrete, and unexpected, progress. Just a decade ago it was not possible to apply machine learning tools on huge data sets, either because data were not of sufficient quality and quantity, or because computing platforms were not powerful enough. If you want a reference date, it is only since 2012 that the deep networks have achieved levels of performance comparable to human ones in tasks of image recognition.

Machine learning and data science are influencing our society and our culture. Man-machine interaction through natural language and data analysis to improve business processes, like recommendation systems to anticipate people's everyday needs, have opened up new forms of market competition. More generally, machines and devices in industrial, business, and consumer settings are achieving a kind of performance that requires new business models. Data analysis and machine learning also have a major impact on the optimization of manufacturing processes. From a purely scientific point of view, the recent success of machine learning has led researchers to think about artificial intelligence in a completely new way.

Although all this is attractive, tangible and effective, there are many aspects of machine learning that need to be developed and understood in greater depth. For example, a complete theory of how deep learning works has not yet been elaborated, and such a theoretical feat would lead to enormous progress. Expanding the spectrum of machine learning applications is the element that will ultimately determine the impact of data science on society.

For Europe's socio-economic system, the challenge is to keep up with this rapidly evolving sector. Otherwise, we would be at a serious disadvantage with respect to competitors around the world. US tech giants, supported by a galaxy of research initiatives and start-up ventures, are industry leaders, as demonstrated for example by the fact that they are ahead of the game in developing self-driving cars. China is also investing enormous resources in machine learning and artificial intelligence. The industry is still in its infancy and the EU can catch up, if it acts now.

Read more about this topic:
Carlo Baldassi. Learning Is a Quantum Question
Daniele Durante. How to Study Dynamic Networks
Dirk Hovy. The algorithm that Prevents Suicide
Antonio Lijoi, Igor Pruenster. Algorithms for Analysis of Complicated Phenomena
Alessia Melegaro. The Network Modeling Chip that Fights Influenza
Raffaella Piccarreta, Marco Bonetti. A (Statistical) Model for Life