The neural architecture of language: Integrative modeling converges on predictive processing DOI Creative Commons
Martin Schrimpf, Idan Blank, Greta Tuckute

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2020, Volume and Issue: unknown

Published: June 27, 2020

Abstract The neuroscience of perception has recently been revolutionized with an integrative modeling approach in which computation, brain function, and behavior are linked across many datasets computational models. By revealing trends models, this yields novel insights into cognitive neural mechanisms the target domain. We here present a first systematic study taking to higher-level cognition: human language processing, our species’ signature skill. find that most powerful ‘transformer’ models predict nearly 100% explainable variance responses sentences generalize different imaging modalities (fMRI, ECoG). Models’ fits (‘brain score’) behavioral both strongly correlated model accuracy on next-word prediction task (but not other tasks). Model architecture appears substantially contribute fit. These results provide computationally explicit evidence predictive processing fundamentally shapes comprehension brain. Significance Language is quintessentially ability. Research long probed functional mind using diverse imaging, behavioral, approaches. However, adequate neurally mechanistic accounts how meaning might be extracted from sorely lacking. Here, we report important step toward addressing gap by connecting recent artificial networks machine learning recordings during processing. up noise levels. Models perform better at predicting next word sequence also measurements – providing

Language: Английский

Meta-learning in natural and artificial intelligence DOI
Jane X. Wang

Current Opinion in Behavioral Sciences, Journal Year: 2021, Volume and Issue: 38, P. 90 - 95

Published: Jan. 25, 2021

Language: Английский

Citations

119
Biological constraints on neural network models of cognitive function DOI
Friedemann Pulvermüller, Rosario Tomasello, Malte R. Henningsen‐Schomers

et al.

Nature reviews. Neuroscience, Journal Year: 2021, Volume and Issue: 22(8), P. 488 - 502

Published: June 28, 2021

Language: Английский

Citations

114
The “Narratives” fMRI dataset for evaluating models of naturalistic language comprehension DOI Creative Commons
Samuel A. Nastase, Yunfei Liu, Hanna Hillman

et al.

Scientific Data, Journal Year: 2021, Volume and Issue: 8(1)

Published: Sept. 28, 2021

The "Narratives" collection aggregates a variety of functional MRI datasets collected while human subjects listened to naturalistic spoken stories. current release includes 345 subjects, 891 scans, and 27 diverse stories varying duration totaling ~4.6 hours unique stimuli (~43,000 words). This data is well-suited for neuroimaging analysis, intended serve as benchmark models language narrative comprehension. We provide standardized accompanied by rich metadata, preprocessed versions the ready immediate use, story with time-stamped phoneme- word-level transcripts. All code are publicly available full provenance in keeping best practices transparent reproducible neuroimaging.

Language: Английский

Citations

104
A Review of Millimeter Wave Device-Based Localization and Device-Free Sensing Technologies and Applications DOI
Anish Shastri, Neharika Valecha, Enver Bashirov

et al.

IEEE Communications Surveys & Tutorials, Journal Year: 2022, Volume and Issue: 24(3), P. 1708 - 1749

Published: Jan. 1, 2022

The commercial availability of low-cost millimeterwave (mmWave) communication and radar devices is starting to improve the adoption such technologies in consumer markets, paving way for large-scale dense deployments fifthgeneration (5G)-and-beyond as well 6G networks. At same time, pervasive mmWave access will enable device localization device-free sensing with unprecedented accuracy, especially respect sub-6 GHz commercial-grade devices. This paper surveys state art device-based using devices, a focus on indoor deployments. We overview key concepts about signal propagation system design, detailing approaches, algorithms applications sensing. Several dimensions are considered, including main objectives, techniques, performance each work, whether they reached an implementation stage, which hardware platforms or software tools were used. analyze theoretical (including processing machine learning), technological, (hardware prototyping) aspects, exposing under-performing missing techniques items towards enabling highly effective human parameters, position, movement, activity vital signs. Among many interesting findings, we observe that systems would greatly benefit from exposes channel information, better integration between standardcompliant initial algorithms, multiple points (APs). Moreover, more advanced requiring zero-initial knowledge environment help simultaneous mapping (SLAM). Machine learning (ML)-based gaining momentum, but still require collection extensive training datasets, do not yet generalize any environment, limiting their applicability. Device-free (i.e., radar-based) have be improved terms of: accuracy detection signs (respiration heart rate) enhanced robustness/generalization capabilities across different environments; moreover, support needed tracking users, automatic creation networks largescale applications. Finally, integrated performing joint communications infancy: practical advancements required add functionalities mmWave-based protocols based orthogonal frequency-division multiplexing (OFDM) multi-antenna technologies.

Language: Английский

Citations

71
Memristor-based neural networks: a bridge from device to artificial intelligence DOI
Zelin Cao, Bai Sun, Guangdong Zhou

et al.

Nanoscale Horizons, Journal Year: 2023, Volume and Issue: 8(6), P. 716 - 745

Published: Jan. 1, 2023

This paper reviews the research progress in memristor-based neural networks and puts forward future development trends.

Language: Английский

Citations

71
Artificial Neuronal Devices Based on Emerging Materials: Neuronal Dynamics and Applications DOI Creative Commons
Hefei Liu,

Yuan Qin,

Hung‐Yu Chen

et al.

Advanced Materials, Journal Year: 2023, Volume and Issue: 35(37)

Published: Jan. 7, 2023

Abstract Artificial neuronal devices are critical building blocks of neuromorphic computing systems and currently the subject intense research motivated by application needs from new technology more realistic brain emulation. Researchers have proposed a range device concepts that can mimic dynamics functions. Although switching physics structures these artificial neurons largely different, their behaviors be described several neuron models in unified manner. In this paper, reports based on emerging volatile materials reviewed perspective demonstrated models, with focus functions implemented exploitation for computational sensing applications. Furthermore, neuroscience inspirations engineering methods to enrich remain networks toward realizing full functionalities biological discussed.

Language: Английский

Citations

53
Physics-Guided, Physics-Informed, and Physics-Encoded Neural Networks and Operators in Scientific Computing: Fluid and Solid Mechanics DOI
Salah A. Faroughi, Nikhil M. Pawar, Célio Fernandes

et al.

Journal of Computing and Information Science in Engineering, Journal Year: 2024, Volume and Issue: 24(4)

Published: Jan. 8, 2024

Abstract Advancements in computing power have recently made it possible to utilize machine learning and deep push scientific forward a range of disciplines, such as fluid mechanics, solid materials science, etc. The incorporation neural networks is particularly crucial this hybridization process. Due their intrinsic architecture, conventional cannot be successfully trained scoped when data are sparse, which the case many engineering domains. Nonetheless, provide foundation respect physics-driven or knowledge-based constraints during training. Generally speaking, there three distinct network frameworks enforce underlying physics: (i) physics-guided (PgNNs), (ii) physics-informed (PiNNs), (iii) physics-encoded (PeNNs). These methods advantages for accelerating numerical modeling complex multiscale multiphysics phenomena. In addition, recent developments operators (NOs) add another dimension these new simulation paradigms, especially real-time prediction systems required. All models also come with own unique drawbacks limitations that call further fundamental research. This study aims present review four (i.e., PgNNs, PiNNs, PeNNs, NOs) used state-of-the-art architectures applications reviewed, discussed, future research opportunities presented terms improving algorithms, considering causalities, expanding applications, coupling solvers.

Language: Английский

Citations

48
The nature and neurobiology of fear and anxiety: State of the science and opportunities for accelerating discovery DOI Creative Commons
Shannon E. Grogans, Eliza Bliss‐Moreau, Kristin A. Buss

et al.

Neuroscience & Biobehavioral Reviews, Journal Year: 2023, Volume and Issue: 151, P. 105237 - 105237

Published: May 18, 2023

Language: Английский

Citations

45
Alignment of brain embeddings and artificial contextual embeddings in natural language points to common geometric patterns DOI Creative Commons
Ariel Goldstein,

Avigail Grinstein-Dabush,

Mariano Schain

et al.

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: March 30, 2024

Contextual embeddings, derived from deep language models (DLMs), provide a continuous vectorial representation of language. This embedding space differs fundamentally the symbolic representations posited by traditional psycholinguistics. We hypothesize that areas in human brain, similar to DLMs, rely on represent To test this hypothesis, we densely record neural activity patterns inferior frontal gyrus (IFG) three participants using dense intracranial arrays while they listened 30-minute podcast. From these fine-grained spatiotemporal recordings, derive for each word (i.e., brain embedding) patient. Using stringent zero-shot mapping demonstrate embeddings IFG and DLM contextual have common geometric patterns. The allow us predict given left-out based solely its geometrical relationship other non-overlapping words Furthermore, show capture geometry better than static embeddings. exposes vector-based code natural processing brain.

Language: Английский

Citations

26
Evaluating large language models in theory of mind tasks DOI Creative Commons
Michał Kosiński

Proceedings of the National Academy of Sciences, Journal Year: 2024, Volume and Issue: 121(45)

Published: Oct. 29, 2024

Eleven large language models (LLMs) were assessed using 40 bespoke false-belief tasks, considered a gold standard in testing theory of mind (ToM) humans. Each task included scenario, three closely matched true-belief control scenarios, and the reversed versions all four. An LLM had to solve eight scenarios single task. Older solved no tasks; Generative Pre-trained Transformer (GPT)-3-davinci-003 (from November 2022) ChatGPT-3.5-turbo March 2023) 20% ChatGPT-4 June 75% matching performance 6-y-old children observed past studies. We explore potential interpretation these results, including intriguing possibility that ToM-like ability, previously unique humans, may have emerged as an unintended by-product LLMs' improving skills. Regardless how we interpret outcomes, they signify advent more powerful socially skilled AI-with profound positive negative implications.

Language: Английский

Citations

24