The cognitive sensor nodes and systems (CogSys) lab aims at cracking the inner workings of cognition for low-footprint adaptive computing. To do so, we embrace the synergies of neuroscience, machine learning / AI, and hardware design, where we combine:

• a bottom-up approach consisting in diving into neuroscience research to identify, and then to exploit, key computational primitives of the brain,
• a top-down approach that builds on the versatility and scalability of modern AI research.

Tackling an interdisciplinary challenge requires a complementary team that can assemble all pieces of the puzzle at multiple scales. Some of our key research areas include:

• AI hardware accelerators (recurrent neural networks, graph neural networks, large language models, etc.),
• neuromorphic engineering and spiking/event-based neural network processors (digital, mixed-signal, in-memory),
• NeuroAI and learning algorithms (cortical microcircuits, approximations of backprop with scalable learning rules that are local in space and time, Bayesian frameworks, continual learning, few-shot learning, etc.),
• extreme-edge computing and on-device learning.

Click on the People tab to meet us!