Encoding in Balanced Networks: Revisiting Spike Patterns and Chaos in Stimulus-Driven Systems

Highly connected recurrent neural networks often produce chaotic dynamics, meaning their precise activity is sensitive to small perturbations. What are the consequences of chaos for how such networks encode streams of temporal stimuli? On the one hand, chaos is a strong source of randomness, suggesting that small changes in stimuli will be obscured by intrinsically generated variability. On the other hand, recent work shows that the type of chaos that occurs in spiking networks can have a surprisingly low-dimensional structure, suggesting that there may be room for fine stimulus features to be precisely resolved. Here we show that strongly chaotic networks produce patterned spikes that reliably encode time-dependent stimuli: using a decoder sensitive to spike times on timescales of 10’s of ms, one can easily distinguish responses to very similar inputs. Moreover, recurrence serves to distribute signals throughout chaotic networks so that small groups of cells can encode substantial information about signals arriving elsewhere. A conclusion is that the presence of strong chaos in recurrent networks need not exclude precise encoding of temporal stimuli via spike patterns.Author Summary: Recurrently connected populations of excitatory and inhibitory neurons found in cortex are known to produce rich and irregular spiking activity, with complex trial-to-trial variability in response to input stimuli. Many theoretical studies found this firing regime to be associated with chaos, where tiny perturbations explode to impact subsequent neural activity. As a result, the precise spiking patterns produced by such networks would be expected to be too fragile to carry any valuable information about stimuli, since inevitable sources of noise such as synaptic failure or ion channel fluctuations would be amplified by chaotic dynamics on repeated trials. In this article we revisit the implications of chaos in input-driven networks and directly measure its impact on evoked population spike patterns. We find that chaotic network dynamics can, in fact, produce highly patterned spiking activity which can be used by a simple decoder to perform input-classification tasks. This can be explained by the presence of low-dimensional, input-specific chaotic attractors, leading to a form of trial-to-trial variability that is intermittent, rather than uniformly random. We propose that chaos is a manageable by-product of recurrent connectivity, which serves to efficiently distribute information about stimuli throughout a network.