05.09.2024
Claire Vernade receives ERC Starting Grant for project on algorithms that adapt to dynamic environments
Claire Vernade, a group leader at our Cluster of Excellence, has been awarded an ERC Starting Grant. Her project ‘ConSequentIAL’ (Continual and Sequential Learning for Artificial Intelligence), is receiving roughly 1.25 million euros funding over a period of five years.
Claire Vernade wants to guarantee that machine learning systems are both robust and adaptable in evolving, real-world environments. Machine learning has already led to impressive developments in areas ranging from language modeling to drug discovery. However, it still lacks important capabilities: “To fully realize the potential of artificial intelligence (AI), we need systems that can autonomously adapt and remain reliable even when the data distribution changes or when faced with novel situations,” explains Vernade.
This project builds on techniques developed in the field of reinforcement learning, which involves an agent – the software entity being trained – navigating an environment to reach predefined goals and being able to learn through trial and error. “My project is focused on creating AI agents that can make intelligent decisions about when and how to collect new data in order to learn about and adapt to new circumstances – an ability known as ‘exploration’ in reinforcement learning.”
Vernade wants to build theoretical foundations to combine this ability with existing machine learning models. This will enable algorithms to progressively integrate new and different data, and search for new solutions. The long-term goal is to create powerful machine learning systems that help science and society solve complex problems.
ERC Starting Grants are awarded by the European Research Council. These highly coveted fellowships support excellent early-career researchers who show potential to become leaders in their fields.