Does EI balance emerge naturally from plasticity rules?

The computational importance of excitation/ inhibition balance (hereby referred to as EI balance) has been explored by many studies: some suggest that networks that are balanced are robust to noise, others suggest efficient coding or even flexibility. It is also often tied to commonly observed phenomenons within the brain as well, such as irregular firing or self organized criticality. However, most of the studies regarding EI balance used methods such as stability-optimized circuits, where the methods enforcing balance are not biologically realistic. Of course, real brains may operate on optimized parameters that are perfected by evolution. More likely, though, is that there is some sort of plasticity rule or homeostatic mechanism that gives rise to EI balance. In this paper by Trapp et al., they explored whether it is possible that a plasticity rule named the flux rule would do just that. Their results show that the resulting network, after an hour, would exhibit asynchornized firing rates of a 100-200 ms time scale, where bursts in the network would effect almost all neurons. Such a state, they checked, was balanced. The result was repeated with different initial conditions, network sizes and simulation time, and the results were consistent. However, interestingly, EI balance was not reached when they switched to a different plasticity rule, the Oja rule. In fact, the Oja rule driven the network towards a maximally inbalanced state. Therefore it might also be important to note that not all plasticity rules result in EI balance.


Author: Pei-Hsien Liu(劉沛弦)


Source: Trapp, Philip, Rodrigo Echeveste, and Claudius Gros. "EI balance emerges naturally from continuous Hebbian learning in autonomous neural networks." Scientific reports 8.1 (2018): 8939.