Usually, continual learning assumes that the data arrives sequentially, and we upgrade the model using incoming examples, but the data may be corrected or deleted.
Several court hearings have started regarding requests to delete data used to teach publicly available LLMs. Hence, the question is whether such a model should be rebuilt every time from scratch or whether some techniques can be proposed to remove the impact of selected data on the model used, i.e., machine unlearning techniques. Other rationales for using these techniques are privacy (e.g., regulations related to GDPR - right to be forgotten) or fairness (e.g., bias removal).
The lecture will discuss selected issues related to machine unlearning, i.e., how to remove the influence of a specified subset of training data from a machine learning model.
Michal Wozniak is currently a professor of computer science at the Department of Systems and Computer Networks, Wroclaw University of Science and Technology, Wroclaw, Poland. He has authored or co-authored more than 350 papers and three books. His research focuses on machine learning, particularly compound classification methods, classifier ensembles, data stream mining, and imbalanced data processing. Prof. Wozniak has been involved in research projects related to machine learning and has been a consultant for several commercial projects for well-known Polish companies and public administration. He received numerous prestigious awards for his scientific achievements, such as the IBM Smarter Planet Faculty Innovation Award (twice), the IEEE Outstanding Leadership Award, and several best paper awards at prestigious conferences.
This talk describes an approach to address the Lifelong Open-ended Learning Autonomy (LOLA) problem in the realm of robotics while, at the same time, seeking how to make cognitive autonomous robots useful to humans. It will present a formalization of the problem and discuss its relation to the general problem of alignment in robotics and AI. To address possible paths to its solution, it will describe the efforts that are currently being made within the PILLAR-Robots European project leveraging the concept of purpose and hierarchical motivational systems as an initial step to explore possible solutions. Several implementation examples will be provided based on a multilevel motivational hierarchy embedded within the e-MDB cognitive architecture in real robots operating in real environments. A discussion of open issues and promising research lines will be provided.
Richard J. Duro is Full Professor of Computer Science and Artificial Intelligence at the University of Coruña and Coordinator of the Integrated Group for Engineering Research at this university since 1999. His research interests focus on intelligent systems and autonomous robotics, and his current work is related to cognitive mechanisms for autonomous robots operating in harsh environments and in the field of open-ended learning. In these fields he has been Principal Investigator of more than 20 publicly funded research projects at European and national levels, in six of them as coordinator of interuniversity teams, including the PILLAR-Robots project of the European Horizon call, which he currently coordinates. He has also been the Principal Investigator of more than 60 industry contracts and has participated in more than 100. As a result of this work, he has published more than 150 articles in international journals and attended and presented papers at more than 170 international and national scientific conferences. He has published or edited 8 books and 16 book chapters, produced 5 patents (3 international), 35 software registrations, and created two spin-off companies that have resulted from the work of his group.
As for his international positioning, he is a Senior Member of IEEE and INNS (International Neural Network Society). At INNS he has been elected member of the Board of Governors in the periods 2014-2016 and 2017-19 and 2020-2022, having served as Vice President for Conferences and Technical Activities of the INNS for 3 years. He has also served as Chair of the IEEE Computer Society Working Group on Smart Measurement Systems for two years and has been a founding member and Co-Chair of the Spanish Chapter of the IEEE Systems Man and Cybernetics Society. In recent years, he has acted as a plenary or guest speaker in 7 prestigious international conferences. He has participated in more than 200 technical or program committees of international conferences and in the organization of 15.
We have seen how Virtual and Augmented Reality have started to transform aviation training and now we can optimize extended Reality (XR), where we have the ability to take the analog world and superimpose digital training artifacts and create a ‘mixed reality’ to enhance aviation training with digital computing headsets. Digital computing headsets are intuitive and offer a natural means of interaction with no computer, wire, or touchscreen. This approach has several practical advantages to connect with real time weather data, hand and eye traking, integrated artifiial Intellignece, and overlay digital objects elements onto real-world crew environments such as simulators. Unlike virtual reality—the user is not isolated from their surroundings so we can create simulated flight environments overlayed onto a flight training device. Pilots see their own hands, interact with co-pilots, and touch physical cockpit controls; all while immersed in virtual airspace. The potential of using XR and graphically rendered content in global aviation training is limitless.
Lori Brown has been affiliated with aviation since 1988 and is passionate about aviation education. She is a fellow of the Royal Aeronautical Society and is a Professor, Researcher and Airline Transport Pilot. She has educated ab-initio cadets for British Airways, KLM, and UAE, as well as pilots for national and international government agencies. Lori is a Professor at Western Michigan University, College of Aviation. She has been recognized worldwide for her work as the former chair for the ICAO Next Generation of Aviation Professionals (NGAP) Outreach program and the International Pilot Training Consortium (IPTC) work-stream with the Royal Aeronautical Society Outreach Program. She is an evaluator for American Council on Education; principal investigator (PI) and Co-I for several funded research projects for The Partnership to Enhance General Aviation Safety, Accessibility and Sustainability (PEGASAS), a Federal Aviation Administration (FAA) Center of Excellence for General Aviation. Her work seeks to enhance aviation safety, education, accessibility, and sustainability by partnering the FAA and a national network of world-class researchers, educators, and industry leaders. Her work has been featured internationally at many international conferences, books, and journal publications. She evaluates military aviation training and occupations for the American Council on Education (ACE). Her dedication and passion for transforming aviation training earned her the WMU Michigan Professor of the Year and the WMU Innovative Teaching Award.