Convolutional neural network models describe the encoding subspace of local circuits in auditory cortex DOI Creative Commons

Jereme C. Wingert,

Satyabrata Parida, Sam Norman-Haignere

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 8, 2024

Abstract Auditory cortex encodes information about nonlinear combinations of spectro-temporal sound features. Convolutional neural networks (CNNs) provide an architecture for generalizable encoding models that can predict time-varying activity evoked by natural sounds with substantially greater accuracy than established models. However, the complexity CNNs makes it difficult to discern computational properties support their improved performance. To address this limitation, we developed a method visualize tuning subspace captured CNN. Single-unit data was recorded using high channel-count microelectrode arrays from primary auditory (A1) awake, passively listening ferrets during presentation large set. A CNN fit data, replicating approaches previous work. measure subspace, dynamic spectrotemporal receptive field (dSTRF) measured as locally linear filter approximating input-output relationship at each stimulus timepoint. Principal component analysis then used reduce very set filters smaller typically requiring 2-10 account 90% dSTRF variance. The projected into neuron, and new model only values. able spike rate nearly accurately full Sensory responses could be plotted in providing compact visualization. This revealed diversity responses, consistent contrast gain control emergent invariance modulation phase. Within local populations, neurons formed sparse representation tiling subspace. Narrow spiking, putative inhibitory showed distinct patterns may reflect position cortical circuit. These results demonstrate conceptual link between establish framework interpretation deep learning-based Significance statement mediates discrimination complex Many have been proposed encoding, varying generality, interpretability, ease fitting. It has determine if/what different functional are study shows two families models, convolutional same properties, important analytical accurate easy straightforward interpret (tuning subspace).

Язык: Английский

Task-specific invariant representation in auditory cortex DOI Creative Commons
Charles R. Heller, Gregory R. Hamersky, Stephen V. David

и другие.

eLife, Год журнала: 2023, Номер 12

Опубликована: Авг. 31, 2023

Categorical sensory representations are critical for many behaviors, including speech perception. In the auditory system, categorical information is thought to arise hierarchically, becoming increasingly prominent in higher-order cortical regions. The neural mechanisms that support this robust and flexible computation remain poorly understood. Here, we studied sound ferret primary non-primary cortex while animals engaged a challenging discrimination task. Population-level decoding of simultaneously recorded single neurons revealed task engagement caused emerge cortex. cortex, general enhancement was not specific task-relevant categories. These findings consistent with mixed selectivity models disentanglement, which early regions build an overcomplete representation world allow downstream brain flexibly selectively read out behaviorally relevant, information.

Язык: Английский

Процитировано

4
Unexpected suppression of neural responses to natural foreground versus background sounds in auditory cortex DOI Creative Commons
Gregory R. Hamersky, Luke A. Shaheen, Mateo López Espejo

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Ноя. 21, 2023

Abstract In everyday hearing, listeners encounter complex auditory scenes containing overlapping sounds that must be grouped into meaningful sources, or streamed, to perceived accurately. A common example of this problem is the perception a behaviorally relevant foreground stimulus (speech, vocalizations) in background noise (environmental, machine noise). Studies using foreground/background contrast have shown high-order areas cortex humans pre-attentively form an enhanced representation over stimulus. Achieving invariant requires identifying and grouping features comprise so they can removed from foreground. To study cortical computations underlying concurrent (BG) (FG) stimuli, we recorded single unit responses (AC) ferrets during presentation natural sound excerpts these two categories. primary secondary AC, found overall suppression when BGs FGs were presented concurrently relative sum same stimuli isolation. Surprisingly, percepts emphasize dynamic FGs, FG suppressed paired BG sound. The degree could explained by spectro-temporal statistics unique each Moreover, systematic degradation decreased as categories became progressively less statistically distinct. strongly units AC presence reveals novel insight how acoustic are encoded at early stages processing.

Язык: Английский

Процитировано

2
Different state-dependence of population codes across cortex DOI Creative Commons
Akhil Bandi, Caroline A. Runyan

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Май 26, 2024

During perceptual decision-making, behavioral performance varies with changes in internal states such as arousal, motivation, and strategy. Yet it is unknown how these affect information coding across cortical regions involved differing aspects of sensory perception decision-making. We recorded neural activity from the primary auditory cortex (AC) posterior parietal (PPC) mice performing a navigation-based sound localization task. then modeled transitions strategies used during task performance. Mice transitioned between three latent decision-making strategies: an 'optimal' state two 'sub-optimal' characterized by choice bias frequent errors. Performance strongly influenced population patterns association but not cortex. Surprisingly, individual PPC neurons was better explained external inputs variables suboptimal than optimal state. Furthermore, shared variability (coupling) strongest In AC, similarly weak all states. Together, findings indicate that more linked to

Язык: Английский

Процитировано

0
Convolutional neural network models describe the encoding subspace of local circuits in auditory cortex DOI Creative Commons

Jereme C. Wingert,

Satyabrata Parida, Sam Norman-Haignere

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 8, 2024

Abstract Auditory cortex encodes information about nonlinear combinations of spectro-temporal sound features. Convolutional neural networks (CNNs) provide an architecture for generalizable encoding models that can predict time-varying activity evoked by natural sounds with substantially greater accuracy than established models. However, the complexity CNNs makes it difficult to discern computational properties support their improved performance. To address this limitation, we developed a method visualize tuning subspace captured CNN. Single-unit data was recorded using high channel-count microelectrode arrays from primary auditory (A1) awake, passively listening ferrets during presentation large set. A CNN fit data, replicating approaches previous work. measure subspace, dynamic spectrotemporal receptive field (dSTRF) measured as locally linear filter approximating input-output relationship at each stimulus timepoint. Principal component analysis then used reduce very set filters smaller typically requiring 2-10 account 90% dSTRF variance. The projected into neuron, and new model only values. able spike rate nearly accurately full Sensory responses could be plotted in providing compact visualization. This revealed diversity responses, consistent contrast gain control emergent invariance modulation phase. Within local populations, neurons formed sparse representation tiling subspace. Narrow spiking, putative inhibitory showed distinct patterns may reflect position cortical circuit. These results demonstrate conceptual link between establish framework interpretation deep learning-based Significance statement mediates discrimination complex Many have been proposed encoding, varying generality, interpretability, ease fitting. It has determine if/what different functional are study shows two families models, convolutional same properties, important analytical accurate easy straightforward interpret (tuning subspace).

Язык: Английский

Процитировано

0