The UvA and world-leading technology company Bosch have agreed to extend their established collaboration with the launch of a new public-private research lab. Delta Lab 2 – the follow-up to the successful collaboration Delta Lab 1 – will focus on the use of artificial intelligence and machine learning for applications in computer vision, generative models and causal learning. Delta Lab 2 will form part of ICAI, the national Innovation Center for AI, headquartered on the Amsterdam Science Park. The lab will be headed by the UvA’s Dr Jan-Willem van de Meent and Prof. Theo Gevers. Dr Eric Nalisnick will be the daily lab manager.
‘For Bosch, collaboration and close exchange with academic institutions is an essential component of our efforts in the development of safe, robust, and explainable AI. By expanding and realigning the previously successful collaboration in the UvA-Bosch Delta Lab, we are realizing our ambition of combining cutting-edge research with high application potential,’ says Michael Fausten, Senior Vice President and Head of the Bosch Center for Artificial Intelligence (BCAI).
In the Delta 2 lab, ten PhD students, one postdoc and one lab manager will work on projects over the next five years with a total budget of €5.2 million. aiming for new research on deep (causal and partial differential equation-based) generative models; certainty and causality in machine learning; and 3D computer vision.
Building on success
Gevers: ‘The collaboration between UvA and Bosch in Delta Lab 1 has been a great success. We want to build on that success with new fundamental research into machine learning and computer vision technologies. We are grateful to Bosch for continuing to invest in fundamental research. We also thank the previous directors and researchers for all their efforts, and we will continue to work enthusiastically on unexplored areas of AI.’
Van de Meent: ‘We are excited to continue our productive collaboration. Working with Bosch is a win win. It not only facilitates uptake of AI innovations in industry, but also provides a wealth of use cases that can inspire new innovations, such as incorporating physical knowledge into models, reasoning about their causal structure, and evaluating the level of confidence that can be attributed to predictions.’
March 8, 2022ICAI PodcastComments Off on #19 with Frans van Ette – the impact of AI and why AI is more trendy than blockchain
In episode 19 of the Snoek op Zolder podcast, host Hennie Huijgens talks with Frans van Ette, director of AI at TNO and chair of the Data Sharing working group of the Netherlands AI Coalition, about federated data versus data lakes, involving SMEs in AI developments, the impact of AI on society, the development towards plug-and-play solutions for AI, European Data Spaces, the future of data sovereignty, the Guide to interoperable data sharing for AI applications of the Data Sharing working group of the NL AIC, ELSA-Labs, why AI is more trendy than blockchain and tips for HBO students who want to use data. (The podcast is in Dutch.)
Corona has made it more difficult for PhD students to find each other, while this group benefits a lot from being part of a community. Evy van Weelden started her PhD in 2020 and only saw her fellow PhD students half a year later in person. ICAI is now organizing its first PhD social meetup. Van Weelden: ‘A PhD is like a maze in which you have to find your way. I feel like I could learn a lot from PhD candidates that are in their third or fourth year.’
MasterMinds Lab is a collaboration between Tilburg University, Fontys Hogescholen, ROC Tilburg, Actemium, CastLab, Interpolis, Marel, MultiSIM BV, Municipality of Tilburg, Port of Rotterdam, Royal Netherlands Air Force, SpaceBuzz, TimeAware and WPG Zwijsen.
The research reported in this study is funded by the MasterMinds project, part of the RegionDeal Mid- and West-Brabant, and is co-funded by the Ministry of Economic Affairs and Municipality of Tilburg.
Working on flight simulations with the Royal Netherlands Air Force and MultiSIM sounds exciting. What amazed you so far?
‘Before I started I had some experience with virtual reality (VR), but when I first tried the flight simulation I was very impressed with how realistic it was. The company MultiSIM models PC-7 aircrafts exactly how they are in real life. Some people are prone to simulator sickness, but I’m not, so it was very fun. You feel very present in that visual environment, which is very important for the motivation of learning. There are different levels why this simulation is so realistic: the sounds, the environment and if you put pressure on the stick or throttle, the response of the aircraft is exactly how it would be in real life.’
What exactly are you researching?
‘My project focuses on neurophysiological indicators of learning in VR flight simulations. I am currently looking at the difference between desktop flight simulators and a VR flight simulators. To what extent does the fidelity of the simulation – so the degree to which the flight simulation resembles a real flight – influence the subjective workload or flight performance of the user and their brain activity? This topic fits in several types of fields, but the main one is neuro-ergonomics. With ergonomics you look at how a person interacts with a system. But neuro-ergonomics is more specific: you’re actually looking in the brain while this person interacts with a system, computer or machine. Once we have established models of the brain activity during training, we can try to predict the learning curve in VR flight simulations. Eventually we want to give neuro-feedback to the user, in the hope that it would increase their learning curve.’
What is it like to do your research with two external partners?
‘There is a lot of communication involved, with the partners, and internally with my supervisors at Tilburg University. And there is a lot of brainstorming. Everyone is enthusiastic and proactive. The meetings with the people from the partners are fun and inspiring. They are intelligent and have a lot of content-related feedback.’
What does the collaboration look like in practice? Do you go there?
‘My past data collection took place at Mindlabs, but for the next studies I plan to use the pilot trainees. That will take place in Soesterberg where the Airforce and Multisim are settled or I go to Woensdrecht in Zeeland where the pilot training takes place.’
You started with a study in neuroscience. How did you get into AI?
‘During my masters I did an internship that considered brain-computer interfaces (BCIs), and after that I knew for sure I wanted to continue with this kind of research. In short, BCIs are AI-driven interfaces that translate brain activity into device commands. In other words, we use AI to make sense of the electrical signals that are measured from the brain. BCIs could be applied to find out whether a person could be cognitive overloaded, which can impact safety, attention, but also learning. Our research concerns learning. We hope that with the use of BCIs in VR, we can increase the learning curve of these pilot trainees. Although BCIs in the field of work are relatively new, a lot of research groups worldwide are working on it right now. But as far as I’m aware, no one is researching the impact of BCIs on the learning curve in VR flight training yet.’
Is a PhD something you have to discover along the way?
‘Yes, it always starts with an idea and then you have to find more information and advice. You have to find out whether your ideas are practical. It takes a long time before you can actually start a study or data collection. There are so many fields, so many devices, so many ideas.’
You started your PhD in the middle of Corona time. How was this?
‘Well, everyone was in the same boat of course. And there were a lot of online meetings. Also meetings where we could interact with other PhD candidates and sometimes even play games, which was nice. When the lockdowns were less restrictive we got to see each other in person and we could really interact. And then another lockdown came. Right now we are starting up again, but we will probably continue to work flexible.’
What role could ICAI play in this for you and other PhD students?
‘The last time I had an in-person meetup at ICAI, at the ICAI day in October, I was able to connect with a lot of people from different levels and fields. We were seated at tables with a certain topic, where we could brainstorm. I got a chance to talk to people from different universities, PhD candidates, postdocs and even professors. It was really nice to learn about other projects within ICAI. I learned as well that there are some projects that involve BCIs.’
Would you like to see more meetups specifically aimed at PhD students within ICAI?
‘Something PhD-specific is always nice to have. As a PhD candidate you have different needs than someone who is a postdoc or beyond. If you’re struggling with something in your project, data analysis for example, or the AI part of machine learning, other PhD students can think along with you and recommend something.’
On Friday, March 11, 2022, ICAI organizes the first Social Meetup for PhD students (invite only). Do you want to get to know your fellow ICAI PhD students? Sign up!
Save the date: The ICAI Day – 2022 Summer edition will take place on Wednesday June 1, 2022!
In this new episode of the AI science podcast Snoek op Zolder, host Hennie Huijgens speaks with Stefan Leijnen, lecturer at the University of Applied Sciences Utrecht and coordinator of the research and innovation working group of the Netherlands AI Coalition, about creativity and AI, and how AI relates to art. They also discuss the role of the Netherlands AI Coalition and the role of the Research and Innovation working group, the opportunities and challenges of AI for SME’s, the arrival of regional AI hubs in which entrepreneurs will collaborate, the position of the Dutch ‘HBO’ in research and practice, Europe’s catching up in AI and the role of ELSA-labs in this. They end with examples of hybrid AI in art applications: the Neural Zoo poster, the work of Nadieh Bremer of visualcinnamon.com, fashion designer Amber Jae Slooten and dancer David Middendorp. (The podcast is in Dutch.)
The modern world is facing a number of converging megatrends: population growth, increasing scarcity of natural resources, and a need for the sustainable production of nutritious food. Through biotechnology DSM develops sustainable products, using nature’s toolbox, such as microorganisms. The AI Lab for Bioscience (AI4b.io) aims to accelerate this innovation process using AI technology. Renger Jellema: ‘More time to use our human creativity is going to be the most important thing we will gain from AI.’
Renger Jellema is program manager of AI4b.io, the ICAI AI Lab for Bioscience, and he is Senior Data Scientist at the Biodata & Translation group at DSM Science & Innovation.
The lab’s first press release stated that you are the first lab in Europe to apply AI to life science and bioproduction. Why hasn’t this been done before?
‘Engineers have already been applying mathematical models in life science for decades, but now there are rapid developments in computing power and breakthroughs in AI. The combination of methods and techniques has become a unique playing field to take biotechnology, process technology, food science and even health and nutrition to the next level.’
Is this approach being used now by other researchers as well?
‘Yes, more and more biotech scientists and engineers worldwide are now launching initiatives similar to what we are doing. What is unique about AI4b.io is that we scale down, from cubic meters to nanoliters and from months to milliseconds, and not follow the more common reverse order which brings scale-up issues. We have defined five lines of research: starting with scheduling in factories, to unit operations, to automated labs, to microbial strain developments and screening to microbial cultures and health relationships in the gut.’
What can AI mean for bioscience?
‘Developments can go much faster. Because a lot of the patterns in data is already stored in the AI-models, researchers can directly go to the core of the problem. And then there will be more time for the researchers to be creative. That’s also how I explain it to colleagues who are a bit hesitant: we free up time to interpret results and come up with novel ideas. Right now about 80 percent of our time goes into managing data and doing things repeatedly.’
What kind of questions can you try to answer with the help of AI that you couldn’t answer before?
‘We want to reduce the cost of innovation while accelerating our development cycles. Mathematical models already play a good role in reducing experimental work by calculating possible scenarios in advance. What we expect is that with the help of AI we can develop better models, leading to so-called Digital Twins of microbes, processes, and factories. At DSM, for example, we produce food and feed ingredients using the process of fermentation. We grow microorganisms on sustainable, plant-derived sources such as sugar and carbohydrates. The microorganisms convert the sugar into valuable products in large steel vessels. Using advanced simulation models, we can then predict the behavior of microorganisms and interaction with their environment in such large vessels. Based on that, we can optimize these processes to become more energy efficient and produce fewer by-products.’
Can you give an example of a typical application?
‘We have developed advanced process models that can be used for large-scale fermentation vessels with a scale of 100 m3 and above. The problem with this is that calculating a few minutes of the behavior of such a vessel quickly takes a few days of computational time on a multi-core computing platform. This makes it impossible to track or monitor the process in real time. For this application, AI can be trained to represent these models – easily speeding up the calculations by a factor of 100 – acting as Digital Twins of the real fermentation vessel. The Digital Twin becomes a sophisticated digital copy of the real process.’
What can this research eventually mean to the world?
‘At DSM, we develop novel ways to produce healthy nutritional ingredients to feed the world in a more sustainable way. The Digital Twins I mentioned before, help us in the development of such processes and products, working for example toward meat alternatives using plant-based material. We combine different protein materials with ingredients such as vitamins and other micronutrients to create food solutions that taste good, have an appealing texture and keep you healthy.’
We have just set up a Launch Pad program to coach PhD students entering the job market. You have been working as a researcher in industry for quite some time. What advice would you give them?
‘Connect with scientists in companies that inspire you. If you get a chance to present your work at a company, seize that opportunity. It’s easy to shy away and stay behind your computer. But know that companies are interested in your research and are willing to help you further. Also, exploring how your research findings can be applied in practice, will improve your thought process.
Personally, I did my PhD in collaboration with Hoogovens, the steel giant now called Tata Steel. I could have stayed behind my computer and emailed them regularly to pick up the samples I needed for my modelling activities. But often I chose to visit the plant and talk to the operators who had to collect the samples. There I saw how difficult it was to take those industrial samples from the extremely hot processes and I learned to understand why the samples were sometimes not that good. As a result, I was able to change the procedures to improve my research. You have to get your hands dirty to get the best insights.’
On February 17, 2022, the AI Lab for Bioscience will talk about their current work during the lunch Meetup of ‘ICAI: The Labs’ on AI for Food in the Netherlands. Want to join? Sign up!
Civic AI Lab’s proposal for the UNESCO’s International Research Centre on Artificial Intelligence (IRCAI) has been rated early stage with great potential and therefore made it to the top 100 projects list of AI solutions worldwide. IRCAI is releasing a list of 100 projects solving problems related to the 17 United Nations Sustainable Development Goals with the application of Artificial Intelligence, from all five geographical regions: Africa, Europe and Americas, Asia and the Pacific, and the Middle East. Civic AI Lab is one of ICAI’s 29 labs.
In episode 17 of Snoek op Zolder Inge Molenaar, associate professor at the Behavioural Science Institute (BSI) of Radboud University and leader of the Adaptive Learning Lab, talks about AI, data and learning analytics in education. She discusses the research, use and the didactical and ethical implications of adaptive learning technologies in human learning.
Kai Lemkes has been a recruiter in the AI domain for ten years. Since a few months he has been a matchmaker within the ICAI Launch Pad program where he coaches PhD students. Lemkes: ‘PhD students have a blind spot when entering the labor market.’
What does the ICAI Launch Pad program look like?
‘After a first introductory meeting with the PhD students, I coach them in how they can best prepare for a job application, how to build a resume, how to present themselves on LinkedIn, et cetera. We evaluate that and then look at how this person can best present themselves and enter the labor market. We can also hold a closing meeting on request. My door is always open.’
Why is there a need for Launch Pad?
‘Many of these PhD students are at a crossroads where they don’t really know what they want next. What I encounter a lot is that students want to stay in the domain they’re already in, purely because they already know it. I recently spoke to a girl who was strongly attached to the research domain. But when I asked her to describe her ideal job, she said that she would prefer to keep improving products, give presentations and a number of things that you see much more in the commercial domain. It is therefore very important to show this group clearly what they actually choose. That’s a blind spot.’
‘The AI domain has exploded in just a few years. At the moment, almost every company I work with – mainly top-500 companies – is investing in AI. For that reason, many young professionals are quickly lured abroad by companies. Foreign companies are sometimes a bit more ‘aggressive’ when it comes to recruiting talent. They proactively approach PhD students and offer them a substantial salary.’
What are Dutch companies not doing well besides less actively recruiting talent?
‘I see the recruitment process go wrong quite often. Candidates have to sell themselves in a very short time and that does not always result in a good match. Based on two or three conversations, it is quite difficult to determine whether someone is a good fit for a company for the long term. Right now, you need the luck to meet someone who likes you. If you’re having a bad day, you’re not going to look good. And especially in the technical domain you will find many specialists who are a bit more introverted or who find it less easy to present themselves, so that they already start such a process quite tense or uncertain.’
How can companies better handle this?
‘It is better to set up a process in which a company really experiences a candidate and to schedule interviews with several people from the company and not just one person. It is also a good idea to let a promising candidate speak with the whole team. Because the demand for AI specialists is so enormous right now, you sometimes see that companies present themselves as super high-tech and that a young professional later finds out that it is not that high-tech at all. Or they find out that there are no other specialists with whom they can consult. And then they can feel terribly alone. Companies need to be honest about what they have to offer.’
What is your solution to this problem?
‘I am developing the digital platform Future Impact. This should become a lively community in which students and young professionals can help each other, have peer-to-peer conversations and give ratings to companies. On this platform, companies can also present themselves and tell what they have to offer as an employer. Virtual appointments can then be scheduled for a first acquaintance. I also want to organize meetups here with people who, as PhD students, have made the step into the commercial world and can coach others in this process.’
How did you get into this job?
‘I really enjoy making matches and connecting people. I like to network and chat. I stumbled into the AI domain by accident, but I really fell in love with it. Such beautiful things happen here; startups working on zero-co2 emissions technologies, for example. AI offers so many possibilities.’
What does ICAI mean to you?
‘I am originally a commercial recruiter, but for ICAI I am really more of a coach. And actually, as I found out again, I think that’s the most wonderful job. In this role as a coach, I enter the conversation with a different intention than as a recruiter. That gives me a great deal of satisfaction. In addition, my network is growing. My highest goal is to get to know the entire AI ecosystem of the Netherlands.’
Kai Lemkes is a matchmaking expert within AI. He is the founder of several matchmaking platforms including Future Impact.
The Dutch Research Council (NWO) has awarded nine Veni grants to researchers involved in groundbreaking AI research. The recipients can use the grants – up to a maximum of 280,000 per researcher – to further develop their research ideas over the next three years.ICAI wants to congratulate these colleagues for their achievements!
The AI recipients of a Veni grant:
Continual Learning under Human Guidance dr. E.T. Nalisnick (M), Universiteit van Amsterdam Artificial intelligence (AI) systems need to adapt to new scenarios. Yet, we must ensure that the new behaviours and skills that they acquire are safe. The researcher will develop AI techniques that allow autonomous systems to adapt but to do so cautiously, under the guidance of a human.
Intelligent interactive natural language systems you can trust and control dr. V. Niculae (M), Universiteit van Amsterdam Artificial intelligence agents are seemingly approaching human performance in natural language tasks like automatic translation and dialogue. However, deployed in the wild, such systems are out of control, learning to produce harmful language even unprompted. Using recent machine learning breakthroughs, the researcher rethinks language generation for trustworthiness and controllability.
Efficient AI with material-based neural networks dr. H.C. Ruiz Euler (M), Universiteit Twente The unprecedented success of artificial intelligence (AI) comes at the price of unsustainable computational costs. This project will research the potential of a novel technology for highly efficient AI hardware: “material-based neural networks”. This technology will enable the next generation of efficient AI systems for edge computing and autonomous systems.
Helping computers say what they mean to say dr. J.D. Groschwitz (M), Universiteit van Amsterdam When a computer talks to us, for example when answering a question, it must translate that answer from its inner computer representation to fluent human language. This project combines linguistics and state-of-the-art machine learning to create a language generation system in which the output text expresses exactly what the computer meant to say.
The life and death of white dwarf binary stars dr. J.C.J. van Roestel (M), Universiteit van Amsterdam Double white dwarf stars are a rare but important type of binary star. They are potential supernova progenitors, some merge to form massive rotating white dwarfs, and they also emit gravitational wave radiation. I will combine data from the Dutch BlackGEM telescope with multiple other telescope surveys and use novel machine learning methods to uncover the population of short-period eclipsing white dwarf binary stars across the entire sky. By comparing the observed population and characteristics with binary population synthesis models, I will determine how these double white dwarfs end their life.
Personalizing radiotherapy with Artificial Intelligence: reducing the toxicity burden for cancer survivors Lisanne van Dijk PhD, University Medical Center Groningen (UMCG) Many head and neck cancer patients suffer from persistent severe toxicities following radiotherapy. As survival rates increase, toxicity reduction has become more pivotal. This project uses Artificial Intelligence techniques to predict toxicity trajectories, which can facilitate personalized decision-support to guide physicians in finding optimal strategies to reduce these severe toxicities.
Towards realistic models for spatiotemporal data dr. K. Kirchner (V), Technische Universiteit Delft Many environmental factors, such as temperature or air pollution, are recorded at several locations and dates. Because of limited computing power, a realistic analysis of the resulting large datasets is often unachievable. This project develops computational approaches which enable efficient accurate data analysis and reliable forecasts for phenomena with uncertainty.
Explainable Artificial Intelligence to unravel genetic architecture of complex traits Gennady Roshchupkin PhD, Erasmus MC Medical Center While we have learned that most diseases have a genetic component, we are still far away from understanding the underlying processes. Using Artificial Intelligence, I will investigate the complex relationship between DNA mutations and human health. This will be the basis for development of novel diagnostic, prognostic and therapeutic tools.
Neural networks for efficient storage and communication of information dr. J. Townsend (M), Universiteit van Amsterdam The brain is an extremely efficient system for storing and communicating information. This research will study the use of artificial neural networks, inspired by the mechanisms in the brain, for data compression, enabling faster internet communication and more efficient storage of computer files.
In episode 16 of Snoek op Zolder Dominique Roest, Kwartiermaker Team Rendement Operationele Informatie at the Dutch Police and member of ICAI’s Advisory Board, tells about the huge increase in data in the police world, the ethical aspects that play an important role with that, why AI is such an important asset for detective work, the TROI-team as a startup within the police, how much fun it is to work for the police, the values of good laws and regulations and the future of AI for investigative work.