To continuously learn new things our brain has to change its internal neural representations that encode information across large neuronal ensembles and multiple brain areas. At the same time, neuronal activity patterns in the brain are often unreliable, multidimensional and highly complex, which makes it challenging to investigate learning in neuronal networks.

To gain a mechanistic/analytical understanding of (deep) network learning our lab uses a two-fold research strategy. As a starting point, we characterise learning-induced changes of neuronal ensemble activity in biological neuronal networks of mice using in vivo population calcium imaging, high-throughput image processing and cutting-edge data analysis methods. In parallel, we develop biologically inspired multi-layer artificial neuronal network models (ANNs) that exhibit similar information processing and storage capabilities as experimentally observed in real biological networks. The process of re-engineering artificial neuronal networks on a very abstract, theoretical level is thereby crucial to understand the fundamental underpinnings that govern learning in biological neuronal networks. Moreover, the ANN reverse-engineering and development process helps us to derive new hypotheses of how individual network components (i.e. lateral inhibition, recurrent connection rate, etc.) reshape information coding in the brain. In parallel, taking inspiration from biological network learning allows us to develop new, sophisticated ANNs with superior learning performance and improved generalisation capabilities.

Facts & Figures 

19 team members: 1 senior researcher, 4 postdocs, 13 PhD students, 1 TA

Annual budget 1.300,000 CHF

Areas of AI research

• Bio-plausible Deep Learning,
• Theoretical Neuroscience,
• Meta-Learning,
• Continual Learning

Areas of AI applications 

• Neuro-robotics,
• Bio-inspired Language Modelling and Vision Tasks

Contact 

• Benjamin F. Grewe, Prof. of Neural Circuits, Systems and Neuroinformatics, Institute of Neuroinformatics UZH/ETH Zurich, bgrewe(at)ethz.ch

(as of December 2021)