Dynamic stimulus representations in adapting neuronal networks

Throughout our everyday experience, we are continuously exposed to dynamic and highly complex streams of multimodal sensory information, which we tend to perceive as a series of discrete and coherently bounded sub-sequences [1]. While these 'perceptual events' [2] are unfolding, active representations of the relevant stimulus features (such as identity, duration, intensity, etc.) are maintained and ought to be sufficiently discernible by the distributed responses of specifically tuned neuronal populations, transiently associated into coherent ensembles [3]. A primary function of cortical microcircuits and a necessary first step toward more specialized information processing thus lies in their ability to acquire and maintain appropriate and reliable representations of time-varying, sequentially patterned stimuli in a self-organized and experience-dependent manner, through targeted functional modifications of various intrinsic and synaptic properties.In this work, we numerically investigate the relations between several important principles of functional neurodynamics, involving distributed processing in inhibition-dominated, sparsely and randomly coupled recurrent networks of LIF neurons, endowed with spike-timing dependent adaptation at excitatory and inhibitory synapses. Because the dynamical features of active stimulus representations are necessarily bound to the current state of the circuit, we start by assessing the impact of plasticity on the characteristics of the different activity regimes exhibited by networks driven by stochastic and unspecific background input, showing that plasticity actively maintains a robust state of asynchronous irregular (AI) activity, closely matching several in vivo recordings in awake, behaving animals. This activity regime is shown to be highly robust to large variations in the control parameters, namely E/I balance and rate of external input.The noisy and stochastic nature of firing activity in the AI regime is shown to be fundamental for the development of stable and reproducible dynamic stimulus representations, due to a better exploration of the network's state space. Networks with the ability to re-balance following the perturbation caused by the input stimulus and to actively maintain the AI regime display less variable (from trial to trial), but higher-dimensional stimulus-driven responses. In contrast, static networks cannot recover from repeated perturbations of E/I balance caused by the patterned external stimulus, leading to an increasingly synchronous network state and, consequently, an increasingly constrained and redundant dynamical space, which is detrimental to an adequate stimulus representation.REFERENCES:[1] Schapiro, A. C., Rogers, T. T., Cordova, N. I., Turk-Browne, N. B., and Botvinick, M. M.(2013), Neural representations of events arise from temporal community structure., Nature Neuroscience, 16, 4, 486 92, doi:10.1038/nn.3331[2] Zacks, J. M., Speer, N. K., Swallow, K. M., Braver, T. S., and Reynolds, J. R. (2007), Event perception: a mind-brain perspective., Psychological Bulletin, 133, 2, 273–93, doi:10.1037/0033-2909.133.2.273[3] Singer, W. (2013), Cortical dynamics revisited., Trends in Cognitive Sciences, 17, 12, 616–26,doi:10.1016/j.tics.2013.09.00