Oscillations and spike statistics in biophysical attractor networks

O s c i l l a t i o n s a n d s p i k e s t a t i s t i c s i n b i o p h y s i c a l a t t r a c t o r n e t w o r k s Abstract The work of this thesis concerns how cortical memories are stored and retrieved. In particular, large-scale simulations are used to investigate the extent to which associative attractor theory is compliant with known physiology and in vivo dynamics. The first question we ask is whether dynamical attractors can be stored in a network with realistic connectivity and activity levels. While attractor networks typically use all-to-all connectivity and display high unit firing rates, cortical connectivity is sparse and neuronal firing rates very low. To investigate this apparent discrepancy, we developed a large-scale model of cortical layers 2/3 adhering to known anatomy and connectivity between various cell types. The long-range connectivity data that is not available, due to the experimental limitations obtaining it, was however estimated. This was done by using known activity levels and total synaptic inputs onto a pyramidal cell, and from this data estimate the number of active inputs a cortical cell typically receives. Using these estimates we demonstrated that attractor memories can be stored and retrieved in biologically realistic networks, operating on psychophysical timescales and displaying firing rate patterns similar to in vivo layer 2/3 cells. This was achieved in the presence of additional complexity such as synaptic depression and cellular adaptation. Since we estimated and reproduced the excitatory and inhibitory input a single cell, embedded in the cortical network, hypothetically receives, the retrieval dynamics was not dependent on the size of the network. We investigated whether the introduction of conduction delays involved in scaling the network to cortical scales would break down the dynamics. However, using anything from 2000 to 22 million cells, similar attractor retrieval dynamics was obtained. Even in a network with a similar spatial extent and number of cells as mouse cortex, attractor transitions were surprisingly sharp and fast. These fast transitions were related to the self-balancing inhibitory and excitatory currents in the network. In order to obtain realistic firing rates in the network, strong feedback inhibition was used. With strong feedback inhibition such balance was dynamically maintained for a wide range of excitation …