Quantum Annealing for Constraint Satisfaction and Constrained Optimization
Philippe Codognet, Computer Science, JFLI - CNRS / Sorbonne University / University of Tokyo, Japan
AI and Music with Knowledge Graphs
Valentina Presutti, University of Bologna, Italy
Outracing Champion Gran Turismo Drivers with Deep Reinforcement Learning
Michael Spranger, Sony AI, Japan
AI Grappling with Things
Aldo Gangemi, University of Bologna, Italy
Quantum Annealing for Constraint Satisfaction and Constrained Optimization
Philippe Codognet
Computer Science, JFLI - CNRS / Sorbonne University / University of Tokyo
Japan
Brief Bio
Philippe Codognet received a Ph.D. in Computer Science from University of Bordeaux-I (France) in 1989. He worked at the central research laboratory of Thomson-CSF (now Thales) in Orsay (France) and then joined INRIA, the French National Research Center in Computer Science, with a sabbatical leave in 1997/8 at Sony Computer Science Laboratory in Paris. Since 1998, he is professor of Computer Science at University Pierre & Marie Curie in Paris (now Sorbonne University). In 2003, he moved to Japan and worked as attaché for science and technology at the French Embassy in Japan (Tokyo). Then, on leave at CNRS (French National Center for Scientific Research), he created and directed a joint Japanese-French Laboratory in Computer Science regrouping CNRS, Sorbonne University, University of Tokyo, Keio University and the National Institute of Informatics (Japan). This laboratory became a CNRS International Research Laboratory (IRL) in January 2012. He then worked as Director of the CNRS Office in Tokyo and as attaché for science and university cooperation at the French Embassy in Singapore.
He is back in Japan since September 2019 as director of the JFLI (Japanese-French Laboratory for Informatics), and invited professor at University of Tokyo and at the National Institute of Informatics.
His main research topics are in the domain of artificial intelligence and focus on combinatorial optimization and constraint programming, high-level programming languages, logic, parallel computing and computer-based music. His current interest lies in Quantum Computing, in particular in quantum annealing for constrained optimization problems.
He has over 120 publications in international conferences and journals, h-index=28, citations > 2900 (Google Scholar).
Abstract
Even if the debate is still going on whether quantum supremacy has already been achieved or not, it is clear that quantum computing will have a profound impact on computer science in the next 10 to 20 years. In the domains of combinatorial optimization and problem solving, which are ubiquitous in AI, the use of quantum computers to solve concrete problems has started to raise tremendous interest, in both the gate-based paradigm with the Quantum Approximate Optimization Algorithm and in the adiabatic computing paradigm with Quantum Annealing (QA). QA is an alternative type of computation in which problems are encoded in quantum Hamiltonians (energy functions) and quantum dynamics is used to find solutions (ground states of minimal energy). QA can be seen as derived from simulated annealing, but taking advantage of the quantum tunneling effect to overcome energy barriers and therefore escape local minima during the computation. Quantum computers such as the D-Wave systems are implementing those ideas in hardware, as well as quantum-inspired devices based on classical electronics such as Fujitsu’s Digital Annealing Unit. From a programming point of view, the use of QA computers for solving combinatorial problems is getting easier by the general adoption of the Quadratic Unconstrained Binary Optimization (QUBO) formalism. QUBO has become a standard input language for all “Ising Machines” developed by D-Wave, NTT, Fujitsu, Hitachi, Toshiba, Fixstars Amplify and NEC.
We will describe in this talk how constraint problems can be encoded in QUBO and solved by QA. After introducing the basic concepts of QUBO and quantum annealing, we will detail the encoding of basic and complex constraints and present QUBO models for well-known constraint satisfaction and constrained optimization problems such as N-queens, Magic Square, Costas Arrays, and the Quadratic Assignment Problem (QAP), together with implementation results on quantum and quantum-inspired hardware.
AI and Music with Knowledge Graphs
Valentina Presutti
University of Bologna
Italy
Brief Bio
Valentina is an Associate Professor of Computer Science at the University of Bologna. She is also an Associate Researcher at the Institute of Cognitive Science and Technologies of CNR and coordinator of STLab. She received her Ph.D in Computer Science at the University of Bologna (2006). Her research interests include AI, Semantic Web and Linked Data, Knowledge Extraction, Empirical Semantics, Social Robotics, Ontology and Knowledge Engineering. She coordinates the EUH2020 project Polifonia (2021-2024). She was responsible for several national and EU projects (e.g. MARIO, IKS, ArCo). During her post-doc she worked in NeOn and created ontologydesignpatterns.org and the series WOP, reference resources for semantic web researchers. She has published +150 peer reviewed articles. She is part of the editorial board of J. of Web Semantics (Elsevier), Data Intelligence (MIT Press), JASIST (Wiley), Intelligenza Artificiale (IOS Press), and of "Semantic Web Studies" (IOS Press). She is co-director of International Semantic Web Research Summer School (ISWS) and has served in organisational and scientific roles for several events. Google Profile - https://scholar.google.com/citations?user=dvNHkAwAAAAJ&hl=en
Abstract
Despite the ethical, commercial, and social issues, and the related ongoing debates, AI applications in the creative domains (e.g., generative AI, music recommendation systems) have reached impressive performance. There are many exploitation scenarios in educational contexts, in supporting the human creative process, in musicological or historical research, etc. Nevertheless, there are issues, sometimes overlooked, that those AI have in common. In particular, they are trained on massive data of arguably untraceable provenance, quality, and validation, so that they cannot express the features and motivations leading to their output. This issue limits their applicability in scenarios that require domain-specific and high-quality data. The main message of this talk is that machine-learning AI cannot go much further without - or to put it nicely, can benefit from - curated and formalized knowledge, e.g., ontologies and knowledge graphs. I will support my position by presenting examples from the Polifonia research project, which is creating large knowledge graphs of musical heritage and experimenting with applications that benefit from them.
Outracing Champion Gran Turismo Drivers with Deep Reinforcement Learning
Michael Spranger
Sony AI
Japan
Brief Bio
Michael Spranger is the COO of Sony AI Inc., Sony’s strategic research and development organization established April 2020. Sony AI’s mission is to “unleash human imagination and creativity with AI.” Michael is a roboticist by training with extensive research experience in fields such as Natural Language Processing, robotics, and foundations of Artificial Intelligence. Michael has published more than 70 papers at top AI conferences such as IJCAI, NeurIPS and others. Concurrent to Sony AI, Michael also holds a Senior Researcher position at Sony Computer Science Laboratories, Inc., and is actively contributing to Sony’s overall AI ethics strategy.
Abstract
Many potential applications of artificial intelligence involve making
real-time decisions in physical systems while interacting with humans.
Automobile racing represents an extreme example of these conditions;
drivers must execute complex tactical manoeuvres to pass or block
opponents while operating their vehicles at their traction limits.
Racing simulations, such as the PlayStation game Gran Turismo,
faithfully reproduce the non-linear control challenges of real race
cars while also encapsulating the complex multi-agent interactions.
Here we describe how we trained agents for Gran Turismo that can
compete with the world’s best e-sports drivers using Reinforcement
Learning. We demonstrate the capabilities of our agent, Gran Turismo
Sophy, by winning a head-to-head competition against four of the
world’s best Gran Turismo drivers. By describing how we trained
championship-level racers, we demonstrate the possibilities and
challenges to control complex dynamical systems in domains where
agents must respect imprecisely defined human norms.
AI Grappling with Things
Aldo Gangemi
University of Bologna
Italy
Brief Bio
Aldo Gangemi is Full Professor at University of Bologna, and Director of the Institute for Cognitive Sciences and Technologies of the Italian National Research Council, where he has co-founded the Semantic Technology Lab (STLab) in 2008. His research focuses on Semantic Technologies as an integration of methods from Knowledge Engineering, Semantic Web, Linked Data, Cognitive Science, and Natural Language Processing. His theoretical interests concentrate upon the representation and discovery of knowledge patterns across data, ontologies, natural language, and cognition, using hybrid symbolic/sub-symbolic methods. Applications domains include Cultural Heritage, Robotics, Medicine, Law, eGovernment, Agriculture and Fishery, and Business. He has published more than 250 papers in international peer-reviewed journals, conferences and books (Scholar H-index=57), and seats as EiC or EB member of international journals (Semantic Web, Web Semantics, Data Semantics, Applied Ontology), conference chair (EKAW2008, WWW2015, ESWC2018/9), and has coordinated research teams in 8 EU projects. He is member of the Board of Directors at IMT School for Advanced Studies Lucca.
Abstract
The last breed of AI is very powerful as an assistant using one modality (characters, pixels, samples, voxels, …). While often approximated or even seriously off, it seems like we can use it in many ways. Foundational criticism stands on its lack of grounding. I present some grounding issues, and discuss what is happening in factoring out meaning as a stratified, simultaneous phenomenon.