Sparse high-dimensional decomposition of non-primary auditory cortical receptive fields

PLoS Computational Biology, Journal Year: 2025, Volume and Issue: 21(1), P. e1012721 - e1012721

Published: Jan. 2, 2025

Characterizing neuronal responses to natural stimuli remains a central goal in sensory neuroscience. In auditory cortical neurons, the stimulus selectivity of elicited spiking activity is summarized by spectrotemporal receptive field (STRF) that relates spectrogram. Though effective characterizing primary responses, STRFs non-primary neurons can be quite intricate, reflecting their mixed selectivity. The complexity hence impedes understanding how acoustic representations are transformed along pathway. Here, we focus on relationship between ferret cortex (A1) and secondary region, dorsal posterior ectosylvian gyrus (PEG). We propose estimating fields PEG with respect well-established high-dimensional computational model primary-cortical representations. These "cortical fields" (CortRF) estimated greedily identify salient features modulating turn related corresponding features. Hence, they provide biologically plausible hierarchical decompositions PEG. Such CortRF analysis was applied speech temporally orthogonal ripple combination (TORC) and, for comparison, A1 responses. CortRFs captured more complex than neurons; moreover, models were predictive (but not A1) speech. Our results thus suggest secondary-cortical computed as sparse combinations facilitate encoding stimuli. Thus, adding representation, account single-unit sounds better bypassing it considering input confirm explicit details presumed organization cortex.

Language: Английский

---

Deep neural networks effectively model neural adaptation to changing background noise and suggest nonlinear noise filtering methods in auditory cortex

NeuroImage, Journal Year: 2022, Volume and Issue: 266, P. 119819 - 119819

Published: Dec. 16, 2022

The human auditory system displays a robust capacity to adapt sudden changes in background noise, allowing for continuous speech comprehension despite environments. However, comprehensive studies characterizing this ability, the computations that underly process are not well understood. first step towards understanding complex is propose suitable model, but classical and easily interpreted model system, spectro-temporal receptive field (STRF), cannot match nonlinear neural dynamics involved noise adaptation. Here, we utilize deep network (DNN) adaptation illustrating its effectiveness at reproducing levels of both individual electrodes cortical population. By closely inspecting model's STRF-like over time, find alters gain shape when adapting change. We show DNN allow it perform adaptive control, while change creates filtering by altering inhibitory region field. Further, models nonprimary cortex also exhibit their excitatory regions, suggesting differences mechanisms along hierarchy. These findings demonstrate capability networks offer new hypotheses about performs enable noise-robust perception real-world, dynamic

Language: Английский

---

Large-scale electrophysiology and deep learning reveal distorted neural signal dynamics after hearing loss

eLife, Journal Year: 2023, Volume and Issue: 12

Published: May 10, 2023

Listeners with hearing loss often struggle to understand speech in noise, even a aid. To better the auditory processing deficits that underlie this problem, we made large-scale brain recordings from gerbils, common animal model for human hearing, while presenting large database of and noise sounds. We first used manifold learning identify neural subspace which is encoded found it low-dimensional dynamics within are profoundly distorted by loss. then trained deep network (DNN) replicate coding without analyzed underlying dynamics. primarily impacts spectral processing, creating nonlinear distortions cross-frequency interactions result hypersensitivity background persists after amplification Our results new focus efforts design improved aids demonstrate power DNNs as tool study central structures.

Language: Английский

---

A sparse code for natural sound context in auditory cortex

Current Research in Neurobiology, Journal Year: 2023, Volume and Issue: 6, P. 100118 - 100118

Published: Nov. 29, 2023

Accurate sound perception can require integrating information over hundreds of milliseconds or even seconds. Spectro-temporal models coding by single neurons in auditory cortex indicate that the majority sound-evoked activity be attributed to stimuli with a few tens milliseconds. It remains uncertain how system integrates about sensory context on longer timescale. Here we characterized long-lasting contextual effects (AC) using diverse set natural stimuli. We measured as difference neuron's response probe following two different sounds. Many AC showed lasting than temporal window traditional spectro-temporal receptive field. The duration and magnitude varied substantially across This diversity formed sparse code neural population encoded wider range contexts any constituent neuron. Encoding model analysis indicates explained local population, suggesting recurrent circuits support representation cortex.

Language: Английский

---

A Guided Tutorial on Modelling Human Event-Related Potentials with Recurrent Neural Networks

Sensors, Journal Year: 2022, Volume and Issue: 22(23), P. 9243 - 9243

Published: Nov. 28, 2022

In cognitive neuroscience research, computational models of event-related potentials (ERP) can provide a means developing explanatory hypotheses for the observed waveforms. However, researchers trained in neurosciences may face technical challenges implementing these models. This paper provides tutorial on recurrent neural network (RNN) ERP waveforms order to facilitate broader use research. To exemplify RNN model usage, P3 component evoked by target and non-target visual events, measured at channel Pz, is examined. Input representations experimental events corresponding labels are used optimize supervised learning paradigm. Linking one input representation with multiple waveform labels, then optimizing minimize mean-squared-error loss, causes output approximate grand-average waveform. Behavior be evaluated as principles underlying generation. Aside from fitting such model, current will also demonstrate how classify hidden units their temporal responses characterize them using principal analysis. Statistical hypothesis testing applied data. focuses presenting modelling approach subsequent analysis outputs how-to format, publicly available data shared code. While relatively less emphasis placed specific interpretations response generation, results initiate some interesting discussion points.

Language: Английский

---

Quantitative models of auditory cortical processing

Hearing Research, Journal Year: 2023, Volume and Issue: 429, P. 108697 - 108697

Published: Jan. 14, 2023

Language: Английский

---

A general theoretical framework unifying the adaptive, transient and sustained properties of ON and OFF auditory neural responses

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 20, 2024

Sounds are temporal stimuli decomposed into numerous elementary components by the auditory nervous system. For instance, a to spectro-temporal transformation modelling frequency decomposition performed cochlea is widely adopted first processing step in today's computational models of neural responses. Similarly, increments and decrements sound intensity (i.e., raw waveform itself or its spectral bands) constitute critical features code, with high behavioural significance. However, despite growing attention scientific community on OFF responses, their relationship transient ON, sustained responses adaptation remains unclear. In this context, we propose new general model, based pair linear filters, named "AdapTrans" that captures both ON unifying easy expand framework. We demonstrate filtering audio cochleagrams AdapTrans permits accurately render known properties measured different mammal species such as dependence stimulus fall time preceding duration. Furthermore, integrating our framework gold standard state-of-the-art machine learning predict from stimuli, following supervised training large compilation electrophysiology datasets (ready-to-deploy PyTorch pre-processed shared publicly), show systematically improves prediction accuracy estimated within cortical areas rat ferret brain. Together, these results motivate use for systems neuroscientists willing increase plausibility performances audition.

Language: Английский

---

A general model unifying the adaptive, transient and sustained properties of ON and OFF auditory neural responses

PLoS Computational Biology, Journal Year: 2024, Volume and Issue: 20(8), P. e1012288 - e1012288

Published: Aug. 2, 2024

Sounds are temporal stimuli decomposed into numerous elementary components by the auditory nervous system. For instance, a to spectro-temporal transformation modelling frequency decomposition performed cochlea is widely adopted first processing step in today’s computational models of neural responses. Similarly, increments and decrements sound intensity (i.e., raw waveform itself or its spectral bands) constitute critical features code, with high behavioural significance. However, despite growing attention scientific community on OFF responses, their relationship transient ON, sustained responses adaptation remains unclear. In this context, we propose new general model, based pair linear filters, named AdapTrans , that captures both ON unifying easy expand framework. We demonstrate filtering audio cochleagrams permits accurately render known properties measured different mammal species such as dependence stimulus fall time preceding duration. Furthermore, integrating our framework gold standard state-of-the-art machine learning predict from stimuli, following supervised training large compilation electrophysiology datasets (ready-to-deploy PyTorch pre-processed shared publicly), show systematically improves prediction accuracy estimated within cortical areas rat ferret brain. Together, these results motivate use for systems neuroscientists willing increase plausibility performances audition.

Language: Английский

---

A sparse code for natural sound context in auditory cortex

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: June 14, 2023

Abstract Accurate sound perception can require integrating information over hundreds of milliseconds or even seconds. Spectro-temporal models coding by single neurons in auditory cortex indicate that the majority sound-evoked activity be attributed to stimuli with a few tens milliseconds. It remains uncertain how system integrates about sensory context on longer timescale. Here we characterized long-lasting contextual effects (AC) using diverse set natural stimuli. We measured as difference neuron’s response probe following two different sounds. Many AC showed lasting than temporal window traditional spectro-temporal receptive field. The duration and magnitude varied substantially across This diversity formed sparse code neural population encoded wider range contexts any constituent neuron. Encoding model analysis indicates explained local population, suggesting recurrent circuits support representation cortex.

Language: Английский

---