Search for: All records

Correlations between the spiking of pairs of neurons are often used to study the brain’s representation of sensory or motor variables and neural circuit function and dysfunction. Previous statistical techniques have shown how time-averaged spike-spike correlations can be predicted by the time-averaged relationships between the individual neurons and the local field potential (LFP). However, spiking and LFP are both nonstationary, and spike-spike correlations have nonstationary structure that cannot be accounted for by time-averaged approaches. Here we develop parametric models that predict spike-spike correlations using a small number of LFP-based predictors, and we then apply these models to the problem of tracking changes in spike-spike correlations over time. Parametric models allow for flexibility in the choice of which LFP recording channels and frequency bands to use for prediction, and coefficients directly indicate which LFP features drive correlated spiking. Here we demonstrate our methods in simulation and test the models on experimental data from large-scale multi-electrode recordings in the mouse hippocampus and visual cortex. In single time windows, we find that our parametric models can be as accurate as previous nonparametric approaches, while also being flexible and interpretable. We then demonstrate how parametric models can be applied to describe nonstationary spike-spike correlations measured in sequential time windows. We find that although the patterns of both cortical and hippocampal spike-spike correlations vary over time, these changes are, at least partially, predicted by models that assume a fixed spike-field relationship. This approach may thus help to better understand how the dynamics of spike-spike correlations are related to functional brain states. Since spike-spike correlations are increasingly used as features for decoding external variables from neural activity, these models may also have the potential to improve the accuracy of adaptive decoders and brain machine interfaces. 

Theories of efficient coding propose that the auditory system is optimized for the statistical structure of natural sounds, yet the transformations underlying optimal acoustic representations are not well understood. Using a database of natural sounds including human speech and a physiologically-inspired auditory model, we explore the consequences of peripheral (cochlear) and mid-level (auditory midbrain) filter tuning transformations on the representation of natural sound spectra and modulation statistics. Whereas Fourier-based sound decompositions have constant time-frequency resolution at all frequencies, cochlear and auditory midbrain filters bandwidths increase proportional to the filter center frequency. This form of bandwidth scaling produces a systematic decrease in spectral resolution and increase in temporal resolution with increasing frequency. Here we demonstrate that cochlear bandwidth scaling produces a frequency-dependent gain that counteracts the tendency of natural sound power to decrease with frequency, resulting in a whitened output representation. Similarly, bandwidth scaling in mid-level auditory filters further enhances the representation of natural sounds by producing a whitened modulation power spectrum (MPS) with higher modulation entropy than both the cochlear outputs and the conventional Fourier MPS. These findings suggest that the tuning characteristics of the peripheral and mid-level auditory system together produce a whitened output representation in three dimensions (frequency, temporal and spectral modulation) that reduces redundancies and allows for a more efficient use of neural resources. This hierarchical multi-stage tuning strategy is thus likely optimized to extract available information and may underlies perceptual sensitivity to natural sounds. 

Abstract Despite thelack of invariance problem(the many‐to‐many mapping between acoustics and percepts), human listeners experience phonetic constancy and typically perceive what a speaker intends. Most models of human speech recognition (HSR) have side‐stepped this problem, working with abstract, idealized inputs and deferring the challenge of working with real speech. In contrast, carefully engineered deep learning networks allow robust, real‐world automatic speech recognition (ASR). However, the complexities of deep learning architectures and training regimens make it difficult to use them to provide direct insights into mechanisms that may support HSR. In this brief article, we report preliminary results from a two‐layer network that borrows one element from ASR,long short‐term memorynodes, which provide dynamic memory for a range of temporal spans. This allows the model to learn to map real speech from multiple talkers to semantic targets with high accuracy, with human‐like timecourse of lexical access and phonological competition. Internal representations emerge that resemble phonetically organized responses in human superior temporal gyrus, suggesting that the model develops a distributed phonological code despite no explicit training on phonetic or phonemic targets. The ability to work with real speech is a major advance for cognitive models of HSR.