The MosAIk team has been created in 2024 as a follow-up of former Bird, K, Orpailleur and Synalp teams.

 

The main objective of MosAIk is to study AI techniques/models/systems and the ways in which they are built. This requires an understanding and expertise in the different aspects of AI. In other words, our goal is to build on our understanding and on the confrontation of the properties of the variety of approaches to AI, of their strengths and weaknesses, of their conditions of application, and of the way they are evolving, to both improve this understanding and progress towards more efficient, more effective and more robust AI systems.

 

To achieve this objective, methodologies deployed within MosAIk are required to consider AI systems and approaches from both the foundational (top-down) and the application (bottom-up) viewpoints. In other words, it is part of the goals of MosAIk, at the same time, to study the characteristics of systems, techniques and models from a theoretical and formal perspective, and to better understand the conditions under which those systems, techniques and models can be used to solve complex, difficult problems from other disciplines. For this reason, research works in MosAIk places themselves within the quadrant formed by two axes: one ranging from the purely theoretically-driven, to the purely application-driven, and another one considering the two main paradigms of AI, i.e. approaches that are centered on explicitly representing knowledge in a automatically processable way (Symbolic AI), and approaches that are based on learning from large amounts of data (Machine Learning). We recognize however that this dichotomy is approximative, and much of the work within MosAIk takes its strength from combining the symbolic and machine learning paradigms, as well as theoretical and applicative goals. 

 

  – Symbolic AI, theory: Within MosAIk, we consider knowledge representation formalisms, often based on logics and including challenging aspects such as uncertainty, temporality or change, and the various reasoning paradigms that they enable (whether they are deductive or hypothetical, e.g. case-based reasoning, analogical reasoning). 

 

  – Symbolic AI, practice: The practical counterpart to Symbolic AI is concerned with the study of efficient ways to build systems relying on the explicit, formal representation of knowledge. This includes the field of knowledge engineering, including ontology engineering, knowledge graphs and the creation of semantic web systems. This also relates to practical considerations regarding the applicability of reasoning processes, including their efficiency and performance. 

 

  – Machine learning and deep learning, theory: Multiple aspects are considered as part of the theory of machine learning within MosAIk, including the mathematical foundations of the properties of machine learning models and algorithms, the characterization of their behaviors, or the methodological aspects of their use. While some of that theoretical work applies to specific, classical machine learning methods such as SVMs or deep learning models, others focus more generally on the methodological aspects of training, refining and applying machine learning models in general. 

 

  – Machine learning and deep learning, practice: On the applicative side of machine learning and deep learning, research within the MosAIK team concern techniques and methods for solving specific categories of problems, including recommendation, text mining, data processing, or generation. Research here is guided by collaborations with experts in specific application domains, bringing challenging problems that require devising new models, approaches and techniques in machine learning and deep learning.