Machine learning (ML) or artificial intelligence (AI) became a field of research already in 1956, but only in recent years AI has started to revolutionize computer science with spectacular breakthroughs in such applications as facial recognition, artificial speech, self-driving cars, and the famous victory over the world’s highest ranked player of the strategic game “go”. Why did it take so long for the AI revolution to take off? Apart from the ever-increasing computer power, new theoretical insights, and improved algorithms, especially the availability of large data sets (e.g. via the internet) for AI training and testing has spurred the recent success of deep neural networks and related powerful ML approaches.
Also within the scientific institutes of FNWI, a wealth of data is continuously produced from all sorts of experimental sources in the different academic research labs. Thanks to automatization, parallelisation, high-throughput setups, high-resolution instruments, and fast networks, “big data” has become a practical issue in a wide range of experimental research projects. Analysis of these large data streams is often a grand challenge. The current surge of AI machinery provides therefore a compelling opportunity to help with the analysis of these scientific data streams. Some scientists in the physics, chemistry, and biology institutes have already started to employ off-the-shelf AI techniques for their research. However, since most researchers have rather limited expertise in AI technology, and the successful application of AI for their specific analysis may require non-trivial adaptation, tuning, or even fundamental development of AI software, collaborations with AI experts would be highly beneficial.
The FNWI Institute of Informatics (IvI) is at the forefront of the scientific research field of AI. IvI houses the Amsterdam Machine Learning Laboratory (AMLab) directed by Prof. Max Welling and also the head office of the Innovative Center for Artificial Intelligence (ICAI) is at the Science Park in Amsterdam, well connected to IvI. ICAI is a national initiative that
connects academic, industrial and governmental partners for AI development and application on various research topics. The organisational structure of ICAI is built from (currently) nine research labs focussed on different themes, each connecting a partner from an industry or societal organisation with an academic counterpart.
The FNWI has decided to start a new initiative, the AI4Science laboratory, which aims to join several scientific domain experts with IvI’s computer scientists into a consortium focused on developing novel AI tools for analysis of large scientific data sets. The underlying research question we aim to answer is: How can we detect, classify, and predict relevant patterns in scientific data if they are hidden within large amount of non-relevant data?
The AI4Science laboratory is centered around data-driven techniques that make use of the recent breakthroughs in machine learning where appropriate combined with model driven approaches that can deal with the specifics of scientific data. The lab will start off with the following five PhD projects:
- Classifying radio phenomena in real time with streaming machine learning (API)
- Unraveling chemical structure-property relationships from two-dimensional liquid chromatography and mass spectrometry (HIMS)
- Graphical models to understand and predict bird migration in Europe (IBED)
- Accelerating gravitational wave signal discoveries and analyses with deep learning (IOP)
- Causal discovery to reveal complex gene regulation networks from vast transcriptome data sets (SILS)
Each project deals with the urgent question of how to recognize and extract the scientifically important information from a large stream of noisy data that is currently generated in research groups within the different FNWI institutes. Although (many) more FNWI research projects have the potential to benefit from AI based data analysis, for the current programme only five projects have been selected, with a maximum of one per institute.