The neuroconnectionist research programme

Nature reviews. Neuroscience, Год журнала: 2023, Номер 24(7), С. 431 - 450

Опубликована: Май 30, 2023

Язык: Английский

---

Information-processing dynamics in neural networks of macaque cerebral cortex reflect cognitive state and behavior

Proceedings of the National Academy of Sciences, Год журнала: 2023, Номер 120(2)

Опубликована: Янв. 5, 2023

One of the essential functions biological neural networks is processing information. This includes everything from sensory information to perceive environment, up motor interact with environment. Due methodological limitations, it has been historically unclear how changes during different cognitive or behavioral states and what extent processed within between network neurons in brain areas. In this study, we leverage recent advances calculation dynamics explore neural-level frontoparietal areas AIP, F5, M1 a delayed grasping task performed by three macaque monkeys. While was high all task, interareal varied widely: During visuomotor transformation, AIP F5 formed reciprocally connected unit, while no present memory period. Movement execution globally across predominance feedback direction. Furthermore, fine-scale structure reconfigured at neuron level response conditions, despite differences overall amount present. These results suggest that dynamically form higher-order units according demand information-processing hierarchically organized level, coarse determining state finer reflecting conditions.

Язык: Английский

---

A virtual rodent predicts the structure of neural activity across behaviors

Nature, Год журнала: 2024, Номер 632(8025), С. 594 - 602

Опубликована: Июнь 11, 2024

Язык: Английский

---

Measuring and modeling the motor system with machine learning

Current Opinion in Neurobiology, Год журнала: 2021, Номер 70, С. 11 - 23

Опубликована: Июнь 8, 2021

The utility of machine learning in understanding the motor system is promising a revolution how to collect, measure, and analyze data. field movement science already elegantly incorporates theory engineering principles guide experimental work, this review we discuss growing use learning: from pose estimation, kinematic analyses, dimensionality reduction, closed-loop feedback, its neural correlates untangling sensorimotor systems. We also give our perspective on new avenues where markerless motion capture combined with biomechanical modeling networks could be platform for hypothesis-driven research.

Язык: Английский

---

Biological neurons act as generalization filters in reservoir computing

Proceedings of the National Academy of Sciences, Год журнала: 2023, Номер 120(25)

Опубликована: Июнь 12, 2023

Reservoir computing is a machine learning paradigm that transforms the transient dynamics of high-dimensional nonlinear systems for processing time-series data. Although was initially proposed to model information in mammalian cortex, it remains unclear how nonrandom network architecture, such as modular cortex integrates with biophysics living neurons characterize function biological neuronal networks (BNNs). Here, we used optogenetics and calcium imaging record multicellular responses cultured BNNs employed reservoir framework decode their computational capabilities. Micropatterned substrates were embed architecture BNNs. We first show response static inputs can be classified linear decoder modularity positively correlates classification accuracy. then timer task verify possess short-term memory several 100 ms finally this property exploited spoken digit classification. Interestingly, BNN-based reservoirs allow categorical learning, wherein trained on one dataset classify separate datasets same category. Such not possible when directly decoded by decoder, suggesting act generalization filter improve performance. Our findings pave way toward mechanistic understanding representation within build future expectations realization physical based

Язык: Английский

---

Task-driven neural network models predict neural dynamics of proprioception

Cell, Год журнала: 2024, Номер 187(7), С. 1745 - 1761.e19

Опубликована: Март 1, 2024

Proprioception tells the brain state of body based on distributed sensory neurons. Yet, principles that govern proprioceptive processing are poorly understood. Here, we employ a task-driven modeling approach to investigate neural code neurons in cuneate nucleus (CN) and somatosensory cortex area 2 (S1). We simulated muscle spindle signals through musculoskeletal generated large-scale movement repertoire train networks 16 hypotheses, each representing different computational goals. found emerging, task-optimized internal representations generalize from synthetic data predict dynamics CN S1 primates. Computational tasks aim limb position velocity were best at predicting activity both areas. Since task optimization develops better during active than passive movements, postulate is top-down modulated goal-directed movements.

Язык: Английский

---

Mapping model units to visual neurons reveals population code for social behaviour

Nature, Год журнала: 2024, Номер 629(8014), С. 1100 - 1108

Опубликована: Май 22, 2024

Abstract The rich variety of behaviours observed in animals arises through the interplay between sensory processing and motor control. To understand these sensorimotor transformations, it is useful to build models that predict not only neural responses input 1–5 but also how each neuron causally contributes behaviour 6,7 . Here we demonstrate a novel modelling approach identify one-to-one mapping internal units deep network real neurons by predicting behavioural changes arise from systematic perturbations more than dozen neuronal cell types. A key ingredient introduce ‘knockout training’, which involves perturbing during training match experiments. We apply this model transformations Drosophila melanogaster males complex, visually guided social 8–11 visual projection at interface optic lobe central brain form set discrete channels 12 , prior work indicates channel encodes specific feature drive particular 13,14 Our reaches different conclusion: combinations neurons, including those involved non-social behaviours, male interactions with female, forming population code for behaviour. Overall, our framework consolidates effects elicited various into single, unified model, providing map stimulus type behaviour, enabling future incorporation wiring diagrams 15 model.

Язык: Английский

---

A neural implementation model of feedback-based motor learning

Nature Communications, Год журнала: 2025, Номер 16(1)

Опубликована: Фев. 20, 2025

Abstract Animals use feedback to rapidly correct ongoing movements in the presence of a perturbation. Repeated exposure predictable perturbation leads behavioural adaptation that compensates for its effects. Here, we tested hypothesis all processes necessary motor may emerge as properties controller adaptively updates policy. We trained recurrent neural network control own output through an error-based signal, which allowed it counteract external perturbations. Implementing biologically plausible plasticity rule based on this same signal enabled learn compensate persistent perturbations trial-by-trial process. The activity changes during learning matched those from populations neurons monkey primary cortex — known mediate both movement correction and task. Furthermore, our model natively reproduced several key aspects studies humans monkeys. Thus, features can arise internal circuit controls feedback.

Язык: Английский

---

From Observed Action Identity to Social Affordances

Trends in Cognitive Sciences, Год журнала: 2021, Номер 25(6), С. 493 - 505

Опубликована: Март 20, 2021

A substantial fraction of neurons in the monkey anterior intraparietal area (AIP) and its human homologue phAIP are selective for observed manipulative actions (OMAs).OMA encode identity actions, up to level semantic representation phAIP.OMA may result from combination two visual signals originating superior temporal sulcus (STS) concerning: (i) body movements: (ii) changes hand/object relationship (action effects).Others' beyond grasping, be specified parietal territories, underpinning 'social affordance' processing selection potential behavioral responses parieto-premotor circuits. Others' cause continuously changing retinal images, making it challenging build neural representations action identity. The putative (phAIP) host (OMAs). neuronal activity both AIP allows a stable readout OMA across formats, but exhibit greater invariance generalize verbs. These properties stem convergence movements; body–object relationship. We propose that evolutionarily preserved mechanisms underlie specification observed-actions motor afforded by them, thereby promoting social behavior. Manual skills hallmark primates, particularly humans. They have made possible most our transformational impact on world, which was driven an expanding network cortical areas primate lineage subserves control [1.Padberg J. et al.Parallel evolution involved skilled hand use.J. Neurosci. 2007; 27: 10106-10115Crossref PubMed Scopus (111) Google Scholar, 2.Kaas J.H. Stepniewska I. Evolution posterior cortex parietal-frontal networks specific primates.J. Comp. Neurol. 2016; 524: 595-608Crossref (53) 3.Borra E. al.The macaque lateral grasping network: substrate generating purposeful actions.Neurosci. Biobehav. Rev. 2017; 75: 65-90Crossref (54) 4.Goldring A.B. Krubitzer L.A. Chapter 26 - mammals: manipulation tool use.in: Kaas Evolutionary Neuroscience. Second Edition. Academic Press, 2020: 627-656Crossref Scholar]. Interestingly, equally well-articulated machinery is required resolve complexity (OMAs) (see Glossary) performed other individuals, because this ability critical importance planning during interaction interindividual coordination [5.Lanzilotto M. al.Neuronal encoding self others' head rotation dorsal prefrontal cortex.Sci. Rep. 7: 8571Crossref (7) 6.Sacheli L.M. al.How task interactivity shapes observation.Cereb. Cortex. 2019; 29: 5302-5314Crossref (5) 7.Ninomiya T. al.A causal role frontal cortico–cortical monitoring.Nat. Commun. 2020; 11: 5233Crossref (1) Indeed, as compared with complex static stimuli, such objects [8.Bao P. map object space inferotemporal cortex.Nature. 583: 103-108Crossref (19) Scholar], faces [9.Chang L. Tsao D.Y. code facial brain.Cell. 169: 1013-1028.e14Abstract Full Text PDF (190) Scholar,10.Freiwald W.A. face processing: cells, areas, networks, models.Curr. Opin. Neurobiol. 60: 184-191Crossref (4) gaze direction [11.Shepherd S.V. al.Mirroring attention cortex.Proc. Natl. Acad. Sci. U. S. A. 2009; 106: 9489-9494Crossref (93) posture [12.Kumar al.Transformation ventral stream body-selective patches.Cereb. 215-229Crossref (8) others inherently dynamic their dynamics essential observer's brain compute identity, despite rapid image. This probably reason why James Gibson claimed 'animals far perception environment presents observer' [13.Gibson J.J. Ecological Approach Visual Perception. Houghton Mifflin, 1979Google Body movements fundamental component 'action'; nonetheless, they represent only one component. In fact, much more than set coordinated movements, since aims produce change subject immersed [14.Bonini al.Neurophysiological bases underlying organization intentional understanding intention.Conscious. Cogn. 2013; 22: 1095-1104Crossref (24) Thus, agent's causes target constitute element almost important movement itself, make predictable terms goal [15.Oram M.W. Perrett D.I. Integration form motion polysensory (STPa) monkey.J. Neurophysiol. 1996; 76: 109-129Crossref (283) 16.Kilner J.M. al.Motor activation prior observation predicted movement.Nat. 2004; 1299-1301Crossref 17.Maranesi al.Mirror neuron context.J. 2014; 34: 14827-14832Crossref (43) types signal, specifying: how unfold; will position or shape object, naturally coexist everyday characterize Both elements crucial. For example, same act branch serve secure while climbing, manipulate grabbing fruits, use hit something someone else: spite body-movement similarity, these clearly different consequences. Similarly, effect moving away can achieved pushing it, throwing kicking similar consequence outside world. Here, we first review evidence signatures OMA-identity coding brain, point node function. then elucidate connectional architecture enables integration main sources information needed identity: hand–object-interaction (i.e., attainment goal). Finally, extension model larger variety classes ones addition AIP, should drive future studies computation non-human brain. Area has long been considered crucial system routing regarding 3D [18.Murata al.Selectivity shape, size, orientation AIP.J. 2000; 83: 2580-2601Crossref 19.Verhoef B.-E. al.Effects microstimulation three-dimensional Categorization.PLoS One. 2015; 10e0136543Crossref (11) 20.Schaffelhofer Scherberger H. Object vision parietal, premotor, cortices.eLife. 5e15278Crossref (47) Scholar] [21.Pani al.Grasping execution single area.J. 26: 2342-2355Crossref (45) Scholar,22.Maeda K. al.Functional manipulation-related mirror responding own action.J. 560-572Crossref (40) [23.Perrett al.Frameworks analysis animate actions.J. Exp. Biol. 1989; 146: 87-113Crossref Scholar,24.Singer Sheinberg D.L. Temporal articulated slow sequences integrated poses.J. 2010; 30: 3133-3145Crossref (0) [25.Gamberini al.Sensory caudal aspect macaque's lobule.Brain Struct. Funct. 2018; 223: 1863-1879PubMed Scholar,26.Breveglieri R. al.Neurons modulated medial cortex.Curr. 1218-1225.e3Abstract (6) regions premotor [27.Maranesi al.Cortical affordances action.Front. Psychol. 5: 538Crossref whereas neighboring inferior convexity were deemed play other's [28.Rozzi lobule monkey: electrophysiological characterization motor, sensory, correlation cytoarchitectonic areas.Eur. 2008; 28: 1569-1588Crossref Scholar,29.Rizzolatti G. goal-directed neuron-based understanding.Physiol. 94: 655-706Crossref (239) Extant focused exclusively graspable exception recent investigations recorded monkeys exemplars [30.Lanzilotto al.Anterior area: hub network.Cereb. 1816-1833Crossref (16) Scholar,31.Lanzilotto al.Stable format-dependent monkey's neurons.Proc. 117: 16596-16605Crossref (3) findings latter demonstrate about OMAs nodes network. What through achieve actions? study [31.Lanzilotto displayed marked selectivity another grooming) among including emotional gestures lip smacking screaming), neutral yawning chewing), stimuli still monkey, animal, landscape) presented screen. study, also tested large dragging, dropping, pulling, pushing, rotating, squeezing) previously used reveal action-identity addition, four resulting postures actor (standing sitting) viewpoints (lateral frontal) (Figure 1A ). results showed 38% at least format, distinct sets exhibiting preference exemplar (or exemplars), tuning presentation format example Figure 1A). However, no fully visual-invariant found. dynamically according multiplicative mixing described images [32.Ratan Murty N.A. Arun S.P. Multiplicative image attributes 115: E3276-E3285Crossref Such decoding early signal viewpoint (50 ms after stimulus onset) actor's (at 100 ms) and, slightly later (150 ms), even format-independent manner. Crucially, accuracy decoded depends upon presence subset units maintain relatively formats considerable rescaling firing rate specificities each (as relationship, if exists, between individual AIP? clustering indicated characterized toward lying table (e.g., dragging) closely linked consequently, segregated those already contact manipulated rolling squeezing, 1B). largely independent combinations 1C), suggesting relationships contribution 1A) recently patients participating brain–machine interface clinical trial, allowing researchers record single-neuron rostral [33.Aflalo shared verbs Adv. 6eabb3984Crossref region include [34.Orban G.A. Functional definitions primates.Proc. 283: 1828Google revealed impressive similarities reported monkeys. First, viewpoint, approximately 20% selective, majority them facilitated response 1D), smaller (about 15%) suppressed humans Second, could tuned any tested, coverage uniform Third, population patients, providing significant latency video onset). evident neurons, although difficult reach firm conclusion based available evidence, plausible common ancestor differently sizable OMA-selective exhibited format-invariant (80% invariant 55% invariant), consistent generalization OMA-discrimination tasks [35.Platonov Orban Action observation: less-explored part higher-order vision.Sci. 6: 36742Crossref thus facilitate recruitment reading uniquely capacity. To summarize, primates macaques) remarkably sulcus, variable degree abstraction order access via written words functional basic raise question what anatomical might be. prevalent portion where influence own-hand feedback overall responsiveness found stronger sector tracers injected three positions along rostro-caudal extent physiologically investigated region. confirmed previous [36.Borra connections area.Cereb. 18: 1094-1111Crossref (302) 2A ) quantitative differences connectivity patterns 2B). particular, selectivity, regions, area, occipito-temporal 2C). Although yet neurophysiological lower bank (STS), known target. discharge when isolation some distance other. Furthermore, STS responded similarly many [22.Maeda Scholar,30.Lanzilotto Scholar]; enable assess consequences hand–object interactions. Importantly, unaffected except rigidity food quality. resulted effects essentially rather itself. location corresponds TEa [37.Seltzer B. Pandya D.N. Afferent architectonics surrounding rhesus monkey.Brain Res. 1978; 149: 1-24Crossref (575) prominent projection targeting mainly part, prevail TEa–AIP pathway (red arrow 2C) likely represents source manipulations. Areas IPa/PGa potentially relevant targeted middle-STS [38.Vangeneugden differentiation using parametric space.Cereb. 19: 593-611Crossref (75) features forelimb portrayed stick figures: body-part deformation, encoded 'snapshot' kinematic features, 'motion' neurons. cells provide rich extracting interactions [39.Sliwa Freiwald dedicated brain.Science. 356: 745-749Crossref (85) Scholar,40.Ong W.S. correlates strategic cooperation monkeys.Nat. 2021; 24: 116-128Crossref constituting key proposed 'third pathway' [41.Pitcher D. Ungerleider L.G. Evidence third specialized perception.Trends 25: 100-110Abstract (2) Thus far, there little view independence body-patch display mostly view-dependent line (such walking bending knee) [42.Jellema Neural perceived bodily categorical frame reference.Neuropsychologia. 2006; 44: 1535-1546Crossref (blue 2C), coherently strong anatomo-functional reviewed preceding text suggests receives convergent 3, Key Figure): TEa. Considering homology [43.Jastorff al.Integration cues biological STS.Neuroimage. 2012; 911-921Crossref scheme extended located occipitotemporal extends into fusiform gyrus Scholar]: contribute caused [44.Wurm M.F. Caramazza Lateral encodes perceptual components abstract sociality.Neuroimage. 202116153Crossref By contrast, split middle gyrus, extend correspond activations [45.Beauchamp M.S. streams manipulable movements.Neuron. 2002; 149-159Abstract (409) Scholar,46.Jastorff Human magnetic resonance imaging reveals separation processing.J. 7315-7329Crossref Why represented primates' cortex? As mentioned previously, monkeys, show either responses; however, active reaching–grasping dark, (not ones) genuine response. On basis, visuomotor doing. mechanism would work alongside affordances. physical Scholar,21.Pani [20.Schaffelhofer Scholar,47.Jeannerod objects: transformation.Trends 1995; 314-320Abstract (1029) Scholar,48.Bonini al.Space-dependent neurons.J. 4108-4119Crossref (67) forming parieto-frontal circuit experimentally established visually guided [49.Gallese V. al.Deficit preshaping muscimol injection cortex.Neuroreport.

Язык: Английский

---

Recurrent neural networks with explicit representation of dynamic latent variables can mimic behavioral patterns in a physical inference task

Nature Communications, Год журнала: 2022, Номер 13(1)

Опубликована: Окт. 4, 2022

Abstract Primates can richly parse sensory inputs to infer latent information. This ability is hypothesized rely on establishing mental models of the external world and running simulations those models. However, evidence supporting this hypothesis limited behavioral that do not emulate neural computations. Here, we test by directly comparing behavior primates (humans monkeys) in a ball interception task large set recurrent network (RNN) with or without capacity dynamically track underlying variables. Humans monkeys exhibit similar patterns. primate pattern best captured RNNs endowed dynamic inference, consistent brain uses inferences support flexible physical predictions. Moreover, our work highlights general strategy for using model systems computational hypotheses higher function.

Язык: Английский

---