Local dendritic balance enables learning of efficient representations in networks of spiking neurons

Abstract

How can neural networks learn to efficiently represent complex and high-dimensional inputs via local plasticity mechanisms? Classical models of representation learning assume that feedforward weights are learned via pairwise Hebbian-like plasticity. Here, we show that pairwise Hebbian-like plasticity works only under unrealistic requirements on neural dynamics and input statistics. To overcome these limitations, we derive from first principles a learning scheme based on voltage-dependent synaptic plasticity rules. Here, recurrent connections learn to locally balance feedforward input in individual dendritic compartments and thereby can modulate synaptic plasticity to learn efficient representations. We demonstrate in simulations that this learning scheme works robustly even for complex high-dimensional inputs and with inhibitory transmission delays, where Hebbian-like plasticity fails. Our results draw a direct connection between dendritic excitatory–inhibitory balance and voltage-dependent synaptic plasticity as observed in vivo and suggest that both are crucial for representation learning.

Emergence of synaptic organization and computation in dendrites

“…Computational modeling proposes several possible benefits of this property of inhibitory synapses. Consistent with experiments ( ), modulating local inhibition can change the shape of the learning rule for excitatory synapses ( ; ; ). Based on the relative timing and strength of the inhibitory input, potentiation and depression of excitatory synapses can be attenuated ( ) or even fully inverted ( ; ).…”

“…Consistent with experiments ( ), modulating local inhibition can change the shape of the learning rule for excitatory synapses ( ; ; ). Based on the relative timing and strength of the inhibitory input, potentiation and depression of excitatory synapses can be attenuated ( ) or even fully inverted ( ; ). This inhibitory control over excitatory plasticity might enable the recalibration of selectivity of excitatory synapses in the visual cortex when input from one eye is lost ( ).…”

“…This inhibitory control over excitatory plasticity might enable the recalibration of selectivity of excitatory synapses in the visual cortex when input from one eye is lost ( ). Alternatively, local inhibition can help to avoid redundancies in the neural code and lead to more efficient and diverse feature representations ( ). Finally, gating excitatory plasticity through local inhibition can allow for the rapid reorganization of excitatory synapses during periods of disinhibition, while keeping excitatory organization stable ( ; ).…”

“…These constraints postulate a model in which inhibitory synapses form a stable backbone around which excitatory synapses cluster ( ). Once the density of excitatory and inhibitory synapses has stabilized, a local plasticity rule can further rearrange synapses and establish synaptic clusters to enable flexible computations and efficient learning ( ; ; ; ) ( ).…”

“…Specific activity-dependent plasticity mechanisms drive this connectivity refinement ( ). The exact form of these plasticity mechanisms and their functional implications are an active area of research ( ; ; ; ; ).…”