Recent Advances and Research Trends in Artificial Intelligence (Online Lectures)

Doctoral students are invited to explore the frontiers of modern science through our specialised lecture series, delivered by distinguished professors from Unite! partner universities. There are three lectures in the series.

Lecture 1

➡️Title: Combinatorial optimization - on leaving the intuitions behind and focusing on understanding. A surprisingly short travel from simple dependencies through subfunction identification, to perfectly precise surrogates.

🗣Speaker: Michał Przewoźniczek (Wroclaw Tech)

📅Date and time of the lecture: May 7, 2026 | 

📓Abstract:

Starting from the early days of Evolutionary Algorithms (EAs), these optimizers are constructed using various intuitions. Some are inspired by observations that are not necessarily easy to explain. Some others (e.g., using surrogates to approximate the optimized function value to limit its evaluation costs) are straightforward, but the mechanism itself assumes the lack of precision. In black-box optimization, we can discover variable dependencies. This knowledge is the key to propose operators designed (and formally proven) to have certain features rather than being based on intuitions. Some recent works use variable dependencies to construct high-precision surrogates that may perfectly represent the optimized function even if it is complex. Such surrogates are no longer an approximation of the optimized function. Frequently, high-precision surrogates may uncover the unknown features of many black-box instances (of various problems). Thus, these tools go far beyond a low-cost approximation of the original problem.

Lecture 2

➡️Title: Using AI for localizing faults 

🗣Speaker: Franz Wotawa (Graz University of Technology) 

📅Date and time of the lecture: June 1, 2026 | 2:00 PM (CET)

📓Abstract: 

There has always been a close proximity between Software Engineering and Artificial Intelligence. In this lecture, we discuss the foundations of Artificial intelligence methods and tools used for localizing faults in systems and software. We show how advanced logic-based reasoning methods can be used to find and explain faulty behavior of systems and their application to fault localization in programs. Furthermore, we discuss the use of current machine learning methods and compare them with the outcome of logic-based reasoning. Finally, we focus on current research questions and gaps for making the approaches feasible in practice.

Lecture 3

➡️Title: Learning with Limited Data: Prior Knowledge and Adaptation in Modern Machine Learning

🗣Speaker: Qi Chen (Aalto University) 

📅Date and time of the lecture: June 9, 2026 | 12:00 PM (CET)

📓Abstract: 

Modern machine learning systems often rely on massive datasets and computational resources. However, many real-world applications require models that can learn efficiently from limited data and adapt to changing environments.

In this lecture, we explore the role of prior knowledge and knowledge transfer in building data-efficient machine learning systems. We discuss theoretical foundations and algorithmic approaches for transferring knowledge across tasks, adapting models to new domains, and enabling continual learning without catastrophic forgetting.

The lecture will also highlight recent research directions on how these principles can support more robust, efficient, and adaptable AI systems, with examples from modern deep learning and foundation models.