The New ELLIS Program: The JKU Spearheads AI Molecular Research in Europe

Be it image recognition software, chatbots, or a browser’s search algorithm, self-learning systems are now an undeniable part of everyday life.

Prof. S. Hochreiter
Prof. S. Hochreiter

Corporate groups and major powers are engaged in a global head-to-head technology race. By launching a particularly ambitious project under the management of Johannes Kepler University Linz, the ELLIS initiative aims to ensure that Europe stays in the race.

In an effort to facilitate tangible research breakthroughs, the “ELLIS Machine Learning for Molecule Discovery Program” connects European hubs that focus on AI research. Coordinated by the ELLIS Unit at the JKU (head: Prof. Sepp Hochreiter), academic experts and scholars at universities (such as Oxford, Cambridge, and ETH Zurich) have joined experts at institutions such as the Berlin Charité, and corporations, such as AstraZeneca, to share expertise, experience, and even computer codes. This is one of two new ELLIS programs that have now been approved.

Enabling Concrete Applications
Prof. Günter Klambauer (Institute for Machine Learning) explains: "Each of the partners is conducting their own AI system research to support a specific purpose in molecular research. Here at the JKU, we are now coordinating the network’s in-depth exchange."

But what does this mean in concrete terms? Klambauer added: "For example, our partners at Microsoft Research in Cambridge put together a program where – with just a few clues - you have to find molecules that have certain biochemical properties. We took this dataset and used it to develop an AI system that can quite accurately solve these problems."

Although it is standard industry practice to publish data sets and program codes, "… that's not the end of the story. The mutual exchange is invaluable, creating an open conversation about what worked and what didn't, as well as supporting direct application approaches." Everyone benefits from a sense of mutual trust that supports shortening the developmental processes and avoiding missteps.

AI pioneer Sepp Hochreiter remarked: "Each ELLIS partner is continuing to work in his or her area of expertise, but if anyone runs into obstacles or needs to solve a general problem, that person can draw on the expertise and experience of scientists throughout Europe."

The “Few Shots” Method is Revolutionizing AI
ELLIS partners are currently working on various issues in areas that include drug development, climate change, the global food crisis and - at the JKU – molecules, along with AI that is capable of predicting the molecules’ effects, properties, and activities. The JKU’s efforts extend further, however, by developing a new “Few-Shot Learning Method”, meaning that instead of feeding each new AI system with large amounts of data from scratch, it builds on existing AIs (and their "learning" and "knowledge"), attempting to effectively apply a prior understanding to address new problems. Klambauer elaborated: "In simpler terms, it means taking a pyramid that is already there and just having to design the small capstone yourself." The approach is much more efficient, saving an enormous number of computing years.

JKU Rector Meinhard Lukas is particularly pleased about the start of the new ELLIS program: “When it comes to molecular research, this program is revolutionary and the JKU is driving force behind this European initiative. We shouldn’t leave technological progress to major powers or corporations, but rather, as a European community of values, we should be helping to shape it. We can be proud of the fact that the JKU is a strong, active part of this initiative.”