Sequential and efficient neural-population coding of complex task information

Koay, Sue Ann; Charles, Adam S.; Thiberge, Stephan Y.; Brody, Carlos D.; Tank, David W.

doi:10.1016/j.neuron.2021.10.020

Cited by 59 publications

(48 citation statements)

References 119 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The finding that saccades are dynamically encoded in post-saccadic activity echoes past reports of choice memories represented as sequences of activity in rodent prefrontal and parietal cortex [109][110][111][112] . Neurons engaged by such choice sequences were found to project to the striatum 110 , an area important for associative learning, thus making them plausible candidates for maintaining eligibility traces.…”

Section: Possible Functions Of Post-saccadic Activitysupporting

confidence: 81%

Prospective and retrospective representations of saccadic movements in primate prefrontal cortex

Calangiu

Kollmorgen

Reppas

et al. 2022

Preprint

View full text Add to dashboard Cite

Dorso-lateral prefrontal cortex is thought to contribute to adaptive behavior by integrating temporally dispersed, behaviorally-relevant factors. Past work has revealed a variety of neural representations preceding actions, which are involved in internal processes like planning, working memory and covert attention. Task-related activity following actions has often been reported, but so far lacks a clear interpretation. We leveraged modified versions of classic oculomotor paradigms and population recordings to show that post-saccadic activity is a dominant signal in dorso-lateral prefrontal cortex that is distinct from pre-saccadic activity. Unlike pre-saccadic activity, post-saccadic activity occurs after each saccade, although its strength and duration are modulated by task context and expected rewards. In contrast to representations preceding actions, which appear to be mixed randomly across neurons, post-saccadic activity results in representations that are highly structured at the single-neuron and population level. Overall, the properties of post-saccadic activity are consistent with those of an action memory, an internal process with a possible role in learning and updating spatial representations.

show abstract

Section: Possible Functions Of Post-saccadic Activitysupporting

confidence: 81%

Prospective and retrospective representations of saccadic movements in primate prefrontal cortex

Calangiu

Kollmorgen

Reppas

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Their receptive fields could be modeled by GPs with spatiotemporal covariance functions [95]; these could be useful for artificial tasks with spatiotemporal stimuli such as movies and multivariate timeseries. Neurons with localized but random temporal responses were found to be compatible with manifold coding in a decision-making task [48]. Our GPs are a complementary approach to traditional sparse coding [63] and efficient coding [5, 16] hypotheses; the connections to these other theories are interesting for future research.…”

Section: Discussionmentioning

confidence: 99%

“…Their receptive fields could be modeled by GPs with spatiotemporal covariance functions [90]; these could be useful for artificial tasks with spatiotemporal stimuli such as movies and multivariate timeseries. Neurons with localized but random temporal responses were found to be compatible with manifold coding in a decision-making task [46].…”

Section: Discussionmentioning

confidence: 99%

Structured random receptive fields enable informative sensory encodings

Pandey

Pachitariu

Brunton

et al. 2021

Preprint

View full text Add to dashboard Cite

The brain must represent the outside world in a way that enables an animal to survive and thrive. In early sensory systems, populations of neurons have a variety of receptive fields that are structured to detect features in input statistics. Alongside this structure, experimental recordings consistently show that these receptive fields also have a great deal of unexplained variability, which has often been ignored in classical models of sensory neurons. In this work, we model neuronal receptive fields as random samples from probability distributions in two sensory modalities, using data from insect mechanosensors and from neurons of mammalian primary visual cortex (V1). In particular, we build generative receptive field models where our random distributions are Gaussian processes with covariance functions that match the second-order statistics of experimental receptive data. We show theoretical results that these random feature neurons effectively perform randomized wavelet transform on the inputs in the temporal domain for mechanosensory neurons and spatial domain for V1 neurons. Such a transformation removes irrelevant components in the inputs, such as high-frequency noise, and boosts the signal. We demonstrate that these random feature neurons produce better learning from fewer training samples and with smaller networks in a variety of artificial tasks. The random feature model of receptive fields provides a unifying, mathematically tractable framework to understand sensory encodings across both spatial and temporal domains.

show abstract

“…One only requires a sufficiently ordered sequence to support a notion of timevarying position x ( t ), and a fluctuating input that modulates dx/dt differently at different x . The first condition is met, for instance, by neural activity that evolves along a timeordered manifold, a topology hypothesized to underlie dynamics in several cortical areas (43), or through a sequence of metastable network attractors (9). To model timing control in such a system one could extend work on input-dependent speed control of recurrent network dynamics (17) by making the input itself (presumably coming from another brain area) dynamic and localizing its slowing and/or accelerating effects to different regions of the sequence.…”

Section: Discussionmentioning

confidence: 99%

Precision motor timing via scalar input fluctuations

Pang

Duffy

Bell

et al. 2022

Preprint

View full text Add to dashboard Cite

Complex motor skills like playing piano require precise timing over long periods, without errors accumulating between subprocesses like the left and right hand movements. While biological models can produce motor-like sequences, how the brain quenches timing errors is not well understood. Motivated by songbirds, where the left and right brain nuclei governing song sequences do not connect but may receive low-dimensional thalamic input, we present a model where timing errors in an autonomous sequence generator are continually corrected by one-dimensional input fluctuations. We show in a spiking neural network model how such input can rapidly correct temporal offsets in a propagating spike pulse, recapitulating the precise timing seen in songbird brains. In a reduced, more general model, we show that such timing correction emerges when the spatial profile of the input over the sequence sufficiently reflects its temporal fluctuations, yielding time-locking attractors that slow advanced sequences and hasten lagging ones, up to the input timescale. Unlike models without fluctuating input, our model predicts anti-correlated durations of adjacent segments of the output sequence, which we verify in recorded zebra finch songs. This work provides a bioplausible picture of how temporal precision could arise in extended motor sequences and generally how low-dimensional input could continuously coordinate time-varying output signals.SignificanceComplex motor skills like playing piano require precision timing over long periods, often among multiple components like left and right muscle groups. Although brain-like network models can produce motor-like outputs, timing regulation is not well understood. We introduce a model, inspired by songbird brains, where imprecise timing in a cortical-like system is corrected by a single thalamic input regulating the sequential propagation, or tempo, of cortical activity. This model illuminates a relation between the input’s spatial structure and temporal variation that lets lagging activity hasten and advanced activity slow, which makes a prediction about output timing that we verify in real birdsong. This work reveals a simple, neuroplausible mechanism that may play a role in precision cortical or motor timing.

show abstract

Sequential and efficient neural-population coding of complex task information

Cited by 59 publications

References 119 publications

Prospective and retrospective representations of saccadic movements in primate prefrontal cortex

Prospective and retrospective representations of saccadic movements in primate prefrontal cortex

Structured random receptive fields enable informative sensory encodings

Precision motor timing via scalar input fluctuations

Contact Info

Product

Resources

About