Thalamocortical architectures for flexible cognition and efficient learning

Trends in Cognitive Sciences, Год журнала: 2024, Номер 28(8), С. 739 - 756

Опубликована: Июнь 17, 2024

The brain exhibits a remarkable ability to learn and execute context-appropriate behaviors. How it achieves such flexibility, without sacrificing learning efficiency, is an important open question. Neuroscience, psychology, engineering suggest that reusing repurposing computations are part of the answer. Here, we review evidence thalamocortical architectures may have evolved facilitate these objectives flexibility efficiency by coordinating distributed computations. Recent work suggests prefrontal cortical networks compute with flexible codes, mediodorsal thalamus provides regularization promote efficient reuse. Thalamocortical interactions resemble hierarchical Bayesian computations, their network implementation can be related existing gating, synchronization, hub theories thalamic function. By reviewing recent findings providing novel synthesis, highlight key research horizons integrating computation, cognition, systems neuroscience.

Язык: Английский

---

Adaptive control of synaptic plasticity integrates micro- and macroscopic network function

Neuropsychopharmacology, Год журнала: 2022, Номер 48(1), С. 121 - 144

Опубликована: Авг. 29, 2022

Synaptic plasticity configures interactions between neurons and is therefore likely to be a primary driver of behavioral learning development. How this microscopic-macroscopic interaction occurs poorly understood, as researchers frequently examine models within particular ranges abstraction scale. Computational neuroscience machine offer theoretically powerful analyses in neural networks, but results are often siloed only coarsely linked biology. In review, we connections these areas, asking how network computations change function diverse features vice versa. We review can controlled at synapses by calcium dynamics neuromodulatory signals, the manifestation changes their impacts specialized circuits. conclude that metaplasticity-defined broadly adaptive control plasticity-forges across scales governing what groups can't learn about, when, ends. The metaplasticity discuss acts co-opting Hebbian mechanisms, shifting properties, routing activity brain systems. Asking operations go awry should also useful for understanding pathology, which address context autism, schizophrenia Parkinson's disease.

Язык: Английский

---

Coding of object location by heterogeneous neural populations with spatially dependent correlations in weakly electric fish

PLoS Computational Biology, Год журнала: 2023, Номер 19(3), С. e1010938 - e1010938

Опубликована: Март 3, 2023

Understanding how neural populations encode sensory stimuli remains a central problem in neuroscience. Here we performed multi-unit recordings from the electrosensory system of weakly electric fish Apteronotus leptorhynchus response to located at different positions along rostro-caudal axis. Our results reveal that spatial dependence correlated activity receptive fields can help mitigate deleterious effects these correlations would otherwise have if they were spatially independent. Moreover, using mathematical modeling, show experimentally observed heterogeneities neurons optimize information transmission as object location. Taken together, our important implications for understanding whose display antagonistic center-surround organization Important similarities between and other systems suggest will be applicable elsewhere.

Язык: Английский

---

Adaptation-induced plasticity in the sensory cortex

Journal of Neurophysiology, Год журнала: 2022, Номер 128(4), С. 946 - 962

Опубликована: Сен. 21, 2022

For medical and fundamental reasons, we need to understand adult brain plasticity at several levels: structural, physiological, behavioral. Historically, has been mostly investigated by weakening or removing sensory inputs. The visual system extensively used because diminishing inputs, i.e., deprivation-induced plasticity, permits more tractable findings. present review is centered on the reverse strategy, imposing a novel stimulus, adaptation-induced plasticity. Adaptation refers constant (milliseconds hours) presentation of nonoptimal stimulus (adapter) within receptive field (RF, spatial area that modulates neuronal firing) neuron under observation. We specifically focus how adaptation impacts tuning neurons with other associated properties. After adaptation, cortical respond robustly adapter (before it typically evokes feeble responses) developing alternate curves outlast time. Here, dendritic structure as foundation, synthesize push-pull mechanism development acquisition following adaptation. then explain these changes apply global level across different regions species short description underlying neurochemical changes. Finally, discuss physiopathological consequences conclude some gaps questions be addressed further comprehend such neuroplasticity.

Язык: Английский

---

Experience-driven development of decision-related representations in the auditory cortex

EMBO Reports, Год журнала: 2024, Номер unknown

Опубликована: Ноя. 11, 2024

Язык: Английский

---

Passive exposure to task-relevant stimuli enhances categorization learning

eLife, Год журнала: 2023, Номер 12

Опубликована: Июль 21, 2023

Learning to perform a perceptual decision task is generally achieved through sessions of effortful practice with feedback. Here, we investigated how passive exposure task-relevant stimuli, which relatively effortless and does not require feedback, influences active learning. First, trained mice in sound-categorization various schedules combining training. Mice that received exhibited faster learning, regardless whether this occurred entirely before training or was interleaved between sessions. We next neural-network models different architectures learning rules the task. Networks use statistical properties stimuli enhance separability data via unsupervised during provided best account behavioral observations. further found that, schedules, there an increased alignment weight updates from training, such few can be as effective long periods consistent our These results provide key insights for design efficient combine both natural artificial systems.

Язык: Английский

---

Beyond gradients: Factorized, geometric control of interference and generalization

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2021, Номер unknown

Опубликована: Ноя. 20, 2021

Interference and generalization, which refer to counter-productive useful interactions between learning episodes, respectively, are poorly understood in biological neural networks. Whereas much previous work has addressed these topics terms of specialized brain systems, here we investigated how rules should impact them. We found that plasticity groups neurons can be decomposed into biologically meaningful factors, with factor geometry controlling interference generalization. introduce a "coordinated eligibility theory" is determined according products subject surprise-based metaplasticity. This model computes directional derivatives loss functions, need not align task gradients, allowing it protect networks against catastrophic facilitate Because the model's structure closely related other rules, independent feedback transmitted, introduces widely-applicable framework for interpreting supervised, reinforcement-based, unsupervised nervous systems.

Язык: Английский

---

Passive exposure to task-relevant stimuli enhances categorization learning

eLife, Год журнала: 2024, Номер 12

Опубликована: Янв. 24, 2024

Learning to perform a perceptual decision task is generally achieved through sessions of effortful practice with feedback. Here, we investigated how passive exposure task-relevant stimuli, which relatively effortless and does not require feedback, influences active learning. First, trained mice in sound-categorization various schedules combining training. Mice that received exhibited faster learning, regardless whether this occurred entirely before training or was interleaved between sessions. We next neural-network models different architectures learning rules the task. Networks use statistical properties stimuli enhance separability data via unsupervised during provided best account behavioral observations. further found that, schedules, there an increased alignment weight updates from training, such few can be as effective long periods consistent our These results provide key insights for design efficient combine both natural artificial systems.

Язык: Английский

---

Not optimal, just noisy: the geometry of correlated variability leads to highly suboptimal sensory coding

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2022, Номер unknown

Опубликована: Март 9, 2022

Abstract The brain represents the world through activity of neural populations. Correlated variability across simultaneously recorded neurons (noise correlations) has been observed cortical areas and experimental paradigms. Many studies have shown that correlated improves stimulus coding compared to a null model with no correlations. However, such results do not shed light on whether populations’ achieves optimal coding. Here, we assess optimality noise correlations in diverse datasets by developing two novel models each unique biological interpretation: uniform factor analysis model. We show datasets, populations leads highly suboptimal performance according these models. demonstrate constraints prevent many subsets from achieving models, subselecting based criteria leaves suboptimal. Finally, subpopulation is exponentially small as function dimensionality. Together, geometry sensory

Язык: Английский

---

Learning-induced reorganization of neuronal subnetworks in the primary sensory cortex

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

Опубликована: Фев. 21, 2023

Abstract Perceptual learning alters the representation of sensory input in primary cortex. Alterations neuronal tuning, correlation structure and population activity across many subcortical cortical areas have been observed previous studies. However, relationships between these different neural correlates - to what extent they are relevant specific perceptual tasks still unclear. In this study, we recorded layer 2/3 populations whisker somatosensory cortex (wS1) using vivo two-photon calcium imaging as mice were trained perform a self-initiated, vibration frequency discrimination task. Individual wS1 neurons displayed learning-induced broadening sensitivity within task-related categories only during task performance, reflecting both learning-and context-dependent enhancement category selectivity. Learning increased signal noise correlations pairs that prefer same stimulus (‘within-pool’), whereas decreased neuron (‘across-pool’). Increased animals resulted less accurate decoding from but did not affect animal’s decision respond stimuli. Importantly, within-pool elevated on trials which generated learned behavioral response. We demonstrate drives formation task-relevant ‘like-to-like’ subnetworks may facilitate execution responses. Significance Statement found plasticity tuning pairwise such become increasingly aligned categories, indicating Category-specific increases induced by active points top-down feedback driver subnetworks.

Язык: Английский

---