Plenery speakers
Prof. Gabriella Pasi, University of Milano-Bicocca
Title: Context Knowledge and Large Language models
Abstract:
In recent years, there has been a rapid advancement in generative AI and its applications, highlighted by the widespread release of ChatGPT on the Web, showcasing both its potential and limitations. Large Language Models are one of the core technologies driving generative AI and are currently being used across a wide range of NLP tasks, including machine translation, conversational agents, and more. However, LLMs still expose certain limitations, notably their inability to fully understand and rely on context knowledge relevant to the specific task.
A typical approach is to make use prompting techniques to guide the generation of text by taking into account the so called “in-context”, without modifying the model’s parameters. Recently, a branch of research is focusing on exploring ways to improve the modeling and control of the process of injecting context (world knowledge) into models. In this talk, I will present some approaches aimed at this goal, and I will discuss the research challenge of creating personal language models—LLMs tailored to a specific user knowledge (such as the user expertise and language knowledge of individual users or specific user groups).
Bio
Gabriella Pasi is Professor at the Department of Informatics, Systems, and Communication of the University of Milano-Bicocca, where she leads the Information and Knowledge Representation, Retrieval, and Reasoning (IKR3) research Lab. Her main research interests are related to Natural Language Processing, in particular to tasks such as Information Retrieval, Personalization in systems for information access, and personal and contextual LLMs. Her research is supported by numerous grants; among her recent recognitions is the 2023 Outstanding Research Contributions Award of the Web Intelligence Consortium. She has served as Program Chair and Senior Area Chair of numerous international conferences, and she is Associate Editor of several international journals. She is co-founder and member of the executive board of the ELLIS Unit of Milano. She is Fellow of ELLIS (European Laboratory for Learning and Intelligent Systems) and Fellow of the Web Intelligence Academy
Prof. Marta Kwiatkowska, University of Oxford and IPI PAN
Title: Adversarial robustness certification for neural networks: progress and challenges
Abstract:
Machine learning solutions are revolutionizing AI, but their instability against adversarial examples – small perturbations to inputs that can drastically change the output – raises concerns about the readiness of this technology for widespread deployment. Using illustrative examples, this lecture will give an overview of certification methodologies currently under development, which aim to provide provable guarantees on safety and robustness of neural network decisions. These draw on convex relaxation, uncertainty quantification and Bayesian methods, and include guarantees against adversarial perturbations and patch attacks, as well as quantitative verification, which aims to quantify the proportion of inputs that satisfy an output specification.
Bio
Marta Kwiatkowska is a Professor at the University of Oxford and Polish Academy of Sciences, and Fellow at Trinity College Oxford. Her area of expertise lies in probabilistic and quantitative verification techniques and the synthesis of correct-by-construction systems from quantitative specifications. She led the development of the probabilistic model checker PRISM, winner of the 2024 ETAPS Test-of-Time Tool Award, which has been used to model and verify numerous case studies across a variety of application domains. Recently, she has been focusing on safety and trust in Artificial Intelligence, with an emphasis on robustness guarantees for machine learning. Her research has been supported by two ERC Advanced Grants, VERIWARE and FUN2MODEL, EPSRC Programme Grant on Mobile Autonomy and EPSRC Prosperity Partnership FAIR. Kwiatkowska won the Royal Society Milner Award, the BCS Lovelace Medal and the Van Wijngaarden Award, and received an honorary doctorate from KTH Royal Institute of Technology in Stockholm. She is a Fellow of the Royal Society, Fellow of ACM, Member of Academia Europea and International Honorary Member of AAAS.
Prof. Krzysztof Krawiec, Poznan University of Technology
Title: Algorithmic Intelligence: The Programmatic Framework for AI
Abstract:
Programming languages allow expressing algorithms in a manner that feels natural to humans and can be interpreted by machines. However, it was only four decades ago that the notion of programs generating other programs began to be given serious consideration by the scientific community. Since then, the field of program synthesis has evolved and matured considerably, primarily as a theoretical branch of computer science with limited practical implications. Recent advancements in AI and the pursuit of AGI have led to a renewed appreciation of the power and universality of the programmatic perspective. In this talk, I will discuss the role of algorithmic representations and program synthesis in the contemporary AI, exploring how this paradigm is revolutionizing the field, bridging the symbolic and subsymbolic worlds, and what its implications and opportunities are. In particular, I will show how algorithmic intelligence can address the many limitations of mainstream AI, such as its unquenchable thirst for data and lack of transparency, especially in scenarios that require abstract reasoning, structured representations, modularity and compositionality.
Bio
Krzysztof Krawiec is a Professor of Computer Science at the Poznan University of Technology in Poland, where he currently serves as the head of the Neurosymbolic Systems Group. His primary research areas include program synthesis, neurosymbolic systems, evolutionary computation, and medical imaging. He has authored over 180 publications on these topics and has received the Fulbright Senior Advanced Research Award, two ACM SIGEVO Impact Awards, and was a visiting professor at the University of California and Massachusetts Institute of Technology. He also served as the general chair of GECCO’21, the largest scientific event in the field of evolutionary computation and as an advisor at the Confederation of Laboratories for Artificial Intelligence in Europe. Krzysztof is also a co-founder of the Center for Artificial Intelligence and Machine learning, part of the Horizon 2020 Foundations of Trustworthy AI project, and an associate editor of Genetic Programming and Evolvable Machines and ACM Transactions on Evolutionary Learning and Optimization. In addition to his academic contributions, Professor Krawiec serves as the Chief AI Officer at Optopol Technology and CTO of Hylomorph Solutions Ltd.