Neurons and Cognition (q-bio.NC)

Abstract: Creating a quantitative theory for the cortex poses several challenges and raises numerous questions. For instance, what are the significant scales of the system? Are they micro, meso or macroscopic? What are the relevant interactions? Are they pairwise, higher order or mean-field? And what are the control parameters? Are they noisy, dissipative or emergent? To tackle these issues, we suggest using an approach similar to the one that has transformed our understanding of the state of matter. This includes identifying invariances in the ensemble dynamics of various neuron functional classes, searching for order parameters that connect important degrees of freedom and distinguish macroscopic system states, and identifying broken symmetries in the order parameter space to comprehend the emerging laws when many neurons interact and coordinate their activation. By utilizing multielectrode and multiscale neural recordings, we measure the scale-invariant balance between excitatory and inhibitory neurons. We also investigate a set of parameters that can assist us in differentiating between various functional system states (such as the wake/sleep cycle) and pinpointing broken symmetries that serve different information processing and memory functions. Furthermore, we identify broken symmetries that result in pathological states like seizures.

Abstract: Developing computational models of neural response is crucial for understanding sensory processing and neural computations. Current state-of-the-art neural network methods use temporal filters to handle temporal dependencies, resulting in an unrealistic and inflexible processing flow. Meanwhile, these methods target trial-averaged firing rates and fail to capture important features in spike trains. This work presents the temporal conditioning spiking latent variable models (TeCoS-LVM) to simulate the neural response to natural visual stimuli. We use spiking neurons to produce spike outputs that directly match the recorded trains. This approach helps to avoid losing information embedded in the original spike trains. We exclude the temporal dimension from the model parameter space and introduce a temporal conditioning operation to allow the model to adaptively explore and exploit temporal dependencies in stimuli sequences in a natural paradigm. We show that TeCoS-LVM models can produce more realistic spike activities and accurately fit spike statistics than powerful alternatives. Additionally, learned TeCoS-LVM models can generalize well to longer time scales. Overall, while remaining computationally tractable, our model effectively captures key features of neural coding systems. It thus provides a useful tool for building accurate predictive computational accounts for various sensory perception circuits.