Domain-specific Schema Reuse Supports Flexible Learning to Learn in Primate Brain DOI Creative Commons
Kuan Tian,

Zhiping Zhao,

Yang Chen

и другие.

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

Опубликована: Окт. 27, 2024

Abstract Prior knowledge accelerates subsequent learning of similarly structured problems - a phenomenon termed “learning to learn” by forming and reusing generalizable neural representations, i.e., the schemas. However, stability-plasticity dilemma, how exploit stable schemas facilitate while remaining flexible towards possible changes, is not well understood. We hypothesize that restricting specific functional, e.g., decision-making, subspace making it orthogonal other subspaces allows brain balance stability plasticity. To test it, we trained three macaques on visuomotor mapping tasks recorded activity in dorsolateral premotor cortex. By delineating decision stimulus subspaces, identified schema-like manifold within only subspace. The reuse significantly facilitated learning. In addition, exhibited trend be subspace, minimizing interference between these two domains. Our results revealed functional domains can preserve useful maintaining orthogonality with allowing for adaptation new environments, thereby resolving dilemma. This finding provides insights into mechanisms underlying brain’s capability learn both fast flexibly, which also inspire more efficient algorithms artificial intelligence systems working open, dynamic environments.

Язык: Английский

Artificial cognition vs. artificial intelligence for next-generation autonomous robotic agents DOI Creative Commons
Giulio Sandini, Alessandra Sciutti, Pietro Morasso

и другие.

Frontiers in Computational Neuroscience, Год журнала: 2024, Номер 18

Опубликована: Март 22, 2024

The trend in industrial/service robotics is to develop robots that can cooperate with people, interacting them an autonomous, safe and purposive way. These are the fundamental elements characterizing fourth fifth industrial revolutions (4IR, 5IR): crucial innovation adoption of intelligent technologies allow development cyber-physical systems , similar if not superior humans. common wisdom intelligence might be provided by AI (Artificial Intelligence), a claim supported more media coverage commercial interests than solid scientific evidence. currently conceived quite broad sense, encompassing LLMs lot other things, without any unifying principle, but self-motivating for success various areas. current view mostly follows purely disembodied approach consistent old-fashioned, Cartesian mind-body dualism, reflected software-hardware distinction inherent von Neumann computing architecture. working hypothesis this position paper road next generation autonomous robotic agents cognitive capabilities requires fully brain-inspired, embodied avoids trap dualism aims at full integration Bodyware Cogniware. We name Artificial Cognition (ACo) ground it Cognitive Neuroscience. It specifically focused on proactive knowledge acquisition based bidirectional human-robot interaction: practical advantage enhance generalization explainability. Moreover, we believe brain-inspired network interactions necessary allowing humans artificial agents, building growing level personal trust reciprocal accountability: clearly missing, although actively sought, AI. ACo work progress take number research threads, some antecedent early attempts define concepts methods. In rest will consider blocks need re-visited unitary framework: principles developmental robotics, methods action representation prospection capabilities, role social interaction.

Язык: Английский

Процитировано

12
Memory Reactivation during Sleep Does Not Act Holistically on Object Memory DOI
Elizabeth M. Siefert,

Sindhuja Uppuluri,

Jianing Mu

и другие.

Journal of Neuroscience, Год журнала: 2024, Номер 44(24), С. e0022242024 - e0022242024

Опубликована: Апрель 11, 2024

Memory reactivation during sleep is thought to facilitate memory consolidation. Most research has examined how of specific facts, objects, and associations benefits their overall retention. However, our memories are not unitary, all features a persist in tandem over time. Instead, transformed, with some strengthened others weakened. Does drive transformation? We leveraged the Targeted Reactivation technique an object category learning paradigm examine this question. Participants (20 female, 14 male) learned three categories novel where each had unique, distinguishing as well shared other members its category. used real-time EEG protocol cue these objects at moments optimized generate events. found that improved for while worsening features, suggesting differentiation process. The results indicate does act holistically on memories, instead supporting transformation enhanced others.

Язык: Английский

Процитировано

7
Structure transfer and consolidation in visual implicit learning DOI Open Access
Dominik Garber, József Fiser

Опубликована: Янв. 24, 2025

Transfer learning, the re-application of previously learned higher-level regularities to novel input, is a key challenge in cognition. While previous empirical studies investigated human transfer learning supervised or reinforcement for explicit knowledge, it unknown whether such occurs during naturally more common implicit and unsupervised and, if so, how related memory consolidation. We compared newly acquired abstract knowledge by extending visual statistical paradigm context. found but with important differences depending on explicitness/implicitness knowledge. Observers acquiring initial could structures immediately. In contrast, observers same amount showed opposite effect, structural interference transfer. However, sleep between phases, observers, while still remaining implicit, switched their behaviour pattern as did. This effect was specific not after non-sleep Our results highlight similarities generalizable relying consolidation restructuring internal representations.

Язык: Английский

Процитировано

1
Rapid context inference in a thalamocortical model using recurrent neural networks DOI Creative Commons
Wei‐Long Zheng, Zhongxuan Wu, Ali Hummos

и другие.

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Сен. 27, 2024

Язык: Английский

Процитировано

6
On the Rational Boundedness of Cognitive Control: Shared Versus Separated Representations DOI
Sebastian Musslick, Andrew Saxe,

Abigail Novick Hoskin

и другие.

Опубликована: Ноя. 16, 2020

One of the most fundamental and striking limitations human cognition appears to be a constraint in number control-dependent processes that can executed at one time. This motivates influential tenets cognitive psychology: control relies on central, limited-capacity processing mechanism imposes seriality processing. Here we provide formally explicit challenge this view. We argue causality is reversed: constraints behavior reflect rational bound mechanisms impose processing, prevent interference arises if two or more tasks engage same representations required perform tasks. use both mathematical numerical analyses shared neural network architectures formal grounding for argument–historically known as "multiple-resource theory"–and demonstrate its ability explain wide range phenomena associated with behavior. Furthermore, need control, arising from by different tasks, reflects optimization trade-off intrinsic architectures: increase learning efficacy representations, versus efficiency parallel (i.e., multitasking) task-dedicated representations. The theory helps frame rigorous, normative approach between automaticity, how relates other principles concerning function, computation generally.

Язык: Английский

Процитировано

30
Signatures of task learning in neural representations DOI
Harsha Gurnani, N. Alex Cayco-Gajic

Current Opinion in Neurobiology, Год журнала: 2023, Номер 83, С. 102759 - 102759

Опубликована: Сен. 12, 2023

Язык: Английский

Процитировано

11
Blocked training facilitates learning of multiple schemas DOI Creative Commons
Andre Beukers, Silvy Collin,

Ross P. Kempner

и другие.

Communications Psychology, Год журнала: 2024, Номер 2(1)

Опубликована: Апрель 9, 2024

We all possess a mental library of schemas that specify how different types events unfold. How are these acquired? A key challenge is learning new schema can catastrophically interfere with old knowledge. One solution to this dilemma use interleaved training learn single representation accommodates schemas. However, another class models posits catastrophic interference be avoided by splitting off representations when large prediction errors occur. differentiating that, according models, prevented even under blocked curricula. conducted series semi-naturalistic experiments and simulations Bayesian neural network compare the predictions made "splitting" versus "non-splitting" hypotheses learning. found better performance in compared curricula, explain results using model incorporates representational response errors. In follow-up experiment, we validated inserting early leads than later Our suggest environments (i.e., curricula) play an important role shaping composition.

Язык: Английский

Процитировано

4
Reconciling shared versus context-specific information in a neural network model of latent causes DOI Creative Commons
Qihong Lu, Tan T. Nguyen, Qiong Zhang

и другие.

Scientific Reports, Год журнала: 2024, Номер 14(1)

Опубликована: Июль 22, 2024

Abstract It has been proposed that, when processing a stream of events, humans divide their experiences in terms inferred latent causes (LCs) to support context-dependent learning. However, shared structure is present across contexts, it still unclear how the “splitting” LCs and learning can be simultaneously achieved. Here, we Latent Cause Network (LCNet), neural network model LC inference. Through learning, naturally stores that tasks weights. Additionally, represents context-specific using context module, controlled by Bayesian nonparametric inference algorithm, which assigns unique vector for each LC. Across three simulations, found LCNet could (1) extract function task while avoiding catastrophic interference, (2) capture human data on curriculum effects schema (3) infer underlying event naturalistic videos daily events. Overall, these results demonstrate computationally feasible approach reconciling scalable from laboratory experiment settings settings.

Язык: Английский

Процитировано

4
Structure transfer and consolidation in visual implicit learning DOI Open Access
Dominik Garber, József Fiser

Опубликована: Март 11, 2025

Transfer learning, the re-application of previously learned higher-level regularities to novel input, is a key challenge in cognition. While previous empirical studies investigated human transfer learning supervised or reinforcement for explicit knowledge, it unknown whether such occurs during naturally more common implicit and unsupervised and, if so, how related memory consolidation. We compared newly acquired abstract knowledge by extending visual statistical paradigm context. found but with important differences depending on explicitness/implicitness knowledge. Observers acquiring initial could structures immediately. In contrast, observers same amount showed opposite effect, structural interference transfer. However, sleep between phases, observers, while still remaining implicit, switched their behaviour pattern as did. This effect was specific not after non-sleep Our results highlight similarities generalizable relying consolidation restructuring internal representations.

Язык: Английский

Процитировано

0
Uncovering Cognitive Taskonomy Through Transfer Learning in Masked Autoencoder-Based fMRI Reconstruction DOI
Youzhi Qu, Junfeng Xia, Xinyao Jian

и другие.

Communications in computer and information science, Год журнала: 2025, Номер unknown, С. 35 - 50

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0