Description du projet
Exploiter la puissance de calcul du cerveau humain
L’utilisation de l’IA et de l’apprentissage automatique est aujourd’hui de plus en plus répandue dans des domaines cruciaux tels que la santé, la finance, les véhicules autonomes et la reconnaissance vocale. Toutefois, les sommes énormes investies dans les approches actuelles de ce domaine en matière d’apprentissage automatique et de calcul neuromorphique sont sérieusement limitées, nécessitant une puissance de calcul toujours croissante et des besoins énergétiques élevés. Pour réaliser une percée dans ce domaine, le projet NEU-ChiP financé par l’UE étudiera comment les cellules souches du cerveau humain cultivées sur une micropuce peuvent apprendre à résoudre des problèmes à partir de données. À l’aide d’une modélisation informatique 3D sophistiquée, un consortium interdisciplinaire procédera à une observation des processus de changements des cellules et de leur plasticité pour permettre un changement majeur dans la technologie de l’apprentissage automatique.
Objectif
The EU and the rest of the world increasingly rely on artificial intelligence (AI) and machine learning (ML) for everyday functioning. Applications range from decision making in areas such as health and finance, face recognition, autonomous vehicle control, speech recognition and interaction with the internet and social media platforms. Estimated annual global spend on ML and AI is $77.6B in 2022 with a business value of $3.9T. However, current deep-learning machines suffer from inherent and difficult limitations: architectures not adaptable, ineffective learning rules, long training times and computing power, making advances unsustainable.
The NeuChiP project will tackle this issue. We will use emerging stem cell technology to make human neuronal networks that self-organise developmentally using the rules that form the brain. Networks will be made of layered cortical structures and hubs, with guided directional network connections and housed in a fabricated assembly. Input will be by patterned light at cells expressing optogenetic actuators, and output recorded via high resolution 3D multielectrode arrays. Intrinsic physiological mechanisms will enable them to undergo plasticity to designated input patterns. NeuChip will surpass the abilities of conventional artificial neural networks by conducting tasks in dynamically changing environments, exploiting the adaptive, complex and exploratory nature of biological human neural systems. To achieve this we have assembled a cross-disciplinary consortium of neuroscientists, stem cell biologists, bioelectronics developers, statistical physicists, together with machine learning and neuromorphic computing experts. We expect that within 15 years NeuChiP technology, using biological learning rules and powerful human-brain-based circuits will lead to novel and widespread advances in machine learning abilities and beyond, leading to a paradigm-shift in AI technology and applications to benefit society.
Champ scientifique
- engineering and technologymechanical engineeringvehicle engineeringautomotive engineeringautonomous vehicles
- natural sciencescomputer and information sciencesinternet
- medical and health sciencesmedical biotechnologycells technologiesstem cells
- natural sciencescomputer and information sciencesartificial intelligencemachine learningdeep learning
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence
Mots‑clés
Programme(s)
Régime de financement
RIA - Research and Innovation actionCoordinateur
B4 7ET Birmingham
Royaume-Uni