Plasticity-induced multistability on fast and slow timescales enables optimal information encoding and spontaneous sequence discrimination

Abstract

Neural circuits exhibit remarkable computational flexibility, enabling adaptive responses to noisy and ever-changing environmental cues. A fundamental question in neuroscience concerns how a wide range of behaviors can emerge from a relatively limited set of underlying biological mechanisms. In particular, the interaction between activities of neuronal populations and plasticity modulation of synaptic connections may endow neural circuits with a variety of functional responses when coordinated over different characteristic timescales. Here, we develop an information-theoretic framework to quantitatively explore this idea. We consider a stochastic model for neural activities that incorporates the presence of a coupled dynamic plasticity and time-varying stimuli. We show that long-term plasticity modulations play the functional role of steering neural activities towards a regime of optimal information encoding. By constructing the associated phase diagram, we demonstrate that either Hebbian or anti-Hebbian plasticity may become optimal strategies depending on how the external input is projected to the target neural populations. Conversely, short-term plasticity enables the discrimination of temporal ordering in sequences of inputs by navigating the emergent multistable attractor landscape. By allowing a degree of variability in external stimuli, we also highlight the existence of an optimal variability for sequence discrimination at a given plasticity strength. In summary, the timescale of plasticity modulation shapes how inputs are represented in neural activities, thereby fundamentally altering the computational properties of the system. Our approach offers a unifying information-theoretic perspective of the role of plasticity, paving the way for a quantitative understanding of the emergence of complex computations in coupled neuronal-synaptic dynamics.