Weakly Supervised and Cautious Learning (WSCL)

This year, the 27th edition of the European Conference on Artificial Intelligence (ECAI) – one of the most prestigious events in the field of artificial intelligence research – was held in Santiago de Compostela. The conference, which was attended by internationally renowned researchers, professionals and academics, provides an important opportunity to discuss the challenges and innovations in this constantly evolving field.

The conference featured the ‘Weakly Supervised and Cautious Learning (WSCL)’ workshop. This workshop provided an important opportunity to explore and address some of the most complex challenges that arise in real-world applications of machine learning (ML).

The workshop, coordinated by a group of experts, was attended by Professor Davide Ciucci, Full Professor in the Department of Computer Science, Systems Engineering and Communication at the University of Milan-Bicocca and Principal Investigator of the ReGAInS project.

In recent years, machine learning has been increasingly applied in various contexts, thanks to its ability to analyse large volumes of data, improve the accuracy of decision-making and reduce the human workload. However, these systems are based on the supervised learning paradigm, which relies on the availability of large quantities of data reliably labelled by experts. The reality, however, is very different from these ideal conditions; data may be incomplete or noisy, supervision may be costly or unreliable, and machine learning models may fail to effectively communicate their uncertainties to users, thereby undermining confidence.

Consequently, there has been growing interest in techniques designed to address these issues, including:

• Uncertainty quantification and cautious learning, where machine learning models communicate their own uncertainties to improve reliability and reduce cognitive biases;
• Weak supervised learning and its variants, such as:
  • Incomplete supervision: only some of the training data is labelled;
  • Inaccurate supervision: the data is labelled in a rough manner;
  • Inaccurate supervision: the labels are not always correct.

At the same time, dialogue has intensified between machine learning and related disciplines, such as the representation of knowledge and uncertainty, human–machine interaction, crowdsourcing and active learning.

The aim of the WSCL workshop is to explore how machine learning and related methods can handle weak supervision and provide more cautious and reliable support in the face of data imperfections.

All of this is perfectly in line with the objectives of the ReGAInS project, which aims precisely to develop Artificial Intelligence technologies that are understandable and transparent.