Programme

While ACAI has a long tradition and is organized bi-annualy, this will be the first edition of ESSAI. The European Summer School on Artificial Intelligence will hopefuly become a tradition and be organized annually. In its organization, ESSAI will follow the model of ESSLLI, the European Summer School in Language, Logic and Information.

 

The longer term ambition is that ESSAI would also be a two-week event, with a larger number of courses. For its first edition, ESSAI will be a one-week event.

 

ESSAI and ACAI 2023 will be organized as a single event with six parallel tracks. One of these tracks will be ACAI-2023, The Advanced Course on Artificial Intelligence. 

ESSAI & ACAI 2023 is planned as an in-person event. It is intended that lecturers will be present in-person in Ljubljana, as well as a sizable audience (200-400 attendees). We also intend to live-stream the lectures and allow for a much larger audience to follow the lectures. In the case of unfavorable epidemiological situation, a hybrid or on-line mode can be easily adopted, given that we are making plans for live streaming (as well as recording) of the lectures.

 

The topic of ACAI-2023 will be Artificial Intelligence for Science. The ACAI-2023 Course will comprise ten half-day tutorials, each consisting of two 90 min sessions. The structure of ACAI-2023 is schematically shown above.

The ESSAI School will have five parallel tracks. Each will comprise four courses, each consisting of five 90 min sessions (one on each day during the week). Thus a total of 20 courses will be included in ESSAI. The structure of ESSAI-2023 is schematically shown below.

The ESSAI program will include courses (foundational, introductory or advanced) on different topics from all areas of artificial intelligence. The selection of the courses will made by the ESSAI PC chairs, together with the program committee that they will form for this purpose.

 

Sašo Džeroski (Jožef Stefan Institute, Ljubljana) and Ljupčo Todorovski (University of Ljubljana) act as program chairs of ACAI-2023 and select its topics and lecturers. The topic of ACAI-2023, Artificial Intelligence for Science, is one of the core areas of their scientific research and they also have extensive experience in organizing prominent and varied scientific events. 

 

For ESSAI-2023, Magdalena Ortiz (Umeå University) is the PC chair, while Brian Logan (Utrecht University) and Sašo Džeroski (Jožef Stefan Institute, Ljubljana) are the co-chairs of ESSAI-2023. In addition, a program committee (PC) of prominent scientists from key branches of AI has also been formed. 

In its development, science has gone through several major paradigms. Chronologically ordered from the past to the present, these include (1) the empirical paradigm, focused on describing natural phenomena, (2) the theoretical paradigm that uses models and generalizes over observations, and (3) the computational paradigm that relies on simulating complex phenomena. We are presently witnessing the reign of the fourth paradigm, data-driven science, which heavily uses machine learning to produce novel scientific discoveries.

 

Machine learning is revolutionizing science by helping to handle the flood of scientific data, reinforcing the reign of data-driven science. Of crucial importance here is the ability to handle large and complex data, at which deep neural networks, including convolutional and graph neural networks, excel. Equally important, however, especially for science, is the ability to explain the models learned from data and their predictions. Explainable AI approaches come to the forefront at this point. Finally, the ability to take into account existing knowledge and relate the newly discovered models to it is also important.

 

Besides machine learning, other artificial intelligence approaches are important for science. Among these, computational scientific discovery most directly addresses the challenge of automating and providing support to the process of generating new scientific knowledge. In computational scientific discovery, data, models, and relations between them are considered, with the focus on uncovering these connections. Laws and models constructed by discovery should not only make accurate predictions, but also provide deeper accounts that are consistent with existing scientific theory: They should also be interpretable, ideally being stated in formalisms used by domain scientists. These three points are distinctive features of science that set it apart from other intellectual pursuits. They also impose constraints on the discovery task that mean traditional techniques for machine learning do not suffice to address it.

 

Representing and storing scientific knowledge, whether we are talking about facts or models, is also of paramount importance for its re-use in different scientific processes, by either human or robot scientists. Here ontologies and semantic technologies (and in general knowledge representation approaches) play a key role. The approaches developed in this area bring us closer to automated science and robot scientists, which not only perform experiments, but go beyond this to automate hypotheses formulation, experiment design and close the loop by forming new hypotheses from experimental results. They also bring us closer to open science, where all products of the scientific process (and not only data) are FAIR (Findable, Accessible, Interoperable, and Reusable).