Unite! researchers will work on an innovative proposal in automatic learning artificial intelligence (AI) based on a neuromorphic approach, inspired by the neural networks of the human brain. The project is led, from the UPC, by Professor Jordi Madrenas, together with Diogo Caetano from the Instituto Superior Técnico (IST/INESC MN) in Lisbon (Portugal) and Pawel Herman from KTH (Sweden).
This project has received funding from the Unite! Seed Fund, and will run until December 2024. The goal is to implement next-generation neuromorphic (cervical-type) AI that overcomes the conventional limitations of AI learning and scalability. The team will coordinate the knowledge of the research groups of the three universities to advance along three lines: device, neuron and system. The project partners are experts in each of these areas.
The initial research involves the combination of memristor devices (resistors with memory) for large-scale networks, configurable spiking neuron hardware and high-level brain models. The project is aligned with the European Green Pact, in research on low-power technologies for edge computing. The collaboration between the partners, with complementary profiles, will enable innovative and complementary research to be carried out in addition to the partners' current projects.
“Better understanding the operation of biological neural systems at multiple scales together with efficient hardware will leverage applications in engineering and health domains, including intelligent,low-power consumption robotics, compact neural prostheses and medical aids for neural diseases.”
The project focuses on neuromorphic computation, similar to that of the brain, a highly interdisciplinary field of great relevance for AI. The three Unite! partners have been collaborating for several years.
The UPC brings to the project its leadership and experience in the implementation of low-power neuromorphic systems in real time. In this sense, the group, led by Jordi Madrenas, has developed a suitable architecture for executing, in real time, the behaviour of spiking neural networks (SNNs), including multiple models and plasticity. For their part, researchers from the IST (Portugal) who are experts in cutting-edge R&D in micro- and nanotechnologies will develop memristor technology and electronics for interfacing with FPGA-based systems used for SNNs. The KTH team (Sweden) focuses on computational modelling of cognitive and perceptual brain phenomena, developing neural brain networks for holistic pattern recognition.
Once the first results have been obtained, those responsible plan to prepare a proposal for a more ambitious project in a 2024 call of the EIC (European Innovation Council) Pathfinder 2024 programme, which supports high-risk, cutting-edge projects that explore, like this Unite! project, new areas and deep tech with the aim of developing potential technologies of the future. It is planned at this stage to involve other research centres and companies in cognitive neuroscience. In order to intensify this line of research, the group plans to promote PhD studies and theses with this focus, promoting attractive topics among potential PhD candidates.