Research has shown that the biases present in our society can also be reflected in AI models or even be amplified by them. This poses risks when using these systems in practice. The BIAS project, funded under the Horizon Europe research program, investigates how this occurs in AI applications in the field of HR and in large language models. The project focuses on how these biases can be detected, how they can be reduced, or how to deal with them. It examines applications in various European languages and takes an interdisciplinary approach, incorporating technical, social science, and legal perspectives.

The climate crisis demands an urgent transition to renewable energy. However, despite the clear necessity, renewables are often viewed as financially risky due to their intermittency and market volatility. Investors hesitate, energy providers face uncertainty, and grid operators struggle to balance supply and demand.

Machine learning (ML) forecasting has the potential to transform the economics of renewable energy. Accurate forecasting helps integrate renewables more effectively, supports electrical grid stability, and informs energy trading decisions, making renewables not only viable but profitable. From predicting demand, supply, and prices to determining optimal battery charging and detecting unexpected maintenance needs, ML forecasting enables better decision-making across the entire energy sector.

However, building high-quality forecasting models is time-consuming and requires significant expertise. Scaling this approach to different markets, energy types, and regulatory environments would require an army of data scientists. By automating key steps like feature engineering, model selection, and hyperparameter tuning, AutoML streamlines the forecasting process, making it more robust and scalable. It also provides guardrails to help less experienced data scientists avoid common pitfalls while giving experts the flexibility to experiment with advanced strategies.

In this talk, Nadya Chernyavskaya will show how automated ML forecasting can accelerate the energy transition, making renewable energy a financially sound investment rather than a risky bet.

Robotics is considered the next frontier of AI – advanced mechanical systems from dogs to humanoids are available in abundance, their movements can be controlled in an adaptive way, computer vision and AI seem to have tackled the crucial perception tasks of object detection and recognition. How close are we, however, to robots helping us with everyday chores, assisting elderly or people in physically straining jobs?

In this talk Prof. Dr. Yulia Sandamirskaya will argue that there are two pathways to develop robots capable to work next to humans. One includes a hybrid approach, combining data-driven, pre-trained models with classical robotic algorithms that use known geometry, dynamics, and control rules. The second one draws deep inspiration from biological neural systems and brains and offers a new computational framework that may change how computers are built and programmed in a radical way.

Prof. Dr. Yulia Sandamirskaya will present the field of neuromorphic computing that has been working on such novel computing systems for decades and envision the future in which such systems enable smart, efficient, and useful robotic helpers.

This afternoon pause uses Square Breathing (4:4:4:4) to calm the nervous system and re-center the mind. It’s especially helpful to reset after a long stretch of mental activity.

Natural catastrophes, such as flood, storms, wildfire and earthquakes represent a leading cause of destruction, fatalities and economic losses worldwide. While large earthquakes are relatively rare compared to the other perils, they are still among the costliest natural disasters. 

One of the mechanisms to mitigate the impact of earthquakes is through risk-transfer, where part of the losses is covered by a third party, typically an insurance or reinsurance company. It's therefore crucial for us, to have a good understanding of the seismic risk and expected losses due to earthquakes in the different regions. 

Science has made huge steps in the last decades. Thanks also to the availability of new and better data, current seismic risk models have become quite sophisticated. However, the larger magnitude, loss-causing earthquakes have return periods of tents to hundreds of years and this keeps the modelling of their expected losses quite challenging. 

In this presentation, I will show the key steps and components of a seismic risk model, how we use data to build and calibrate it and how this informs the re-insurance business.