检索结果-内蒙古大学图书馆

THE TEMPORAL STRUCTURE OF LANGUAGE PROCESSING IN THE HUMAN brain CORRESPONDS TO THE LAYERED HIERARCHY OF DEEP LANGUAGE MODELS

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Goldstein, Ariel Ham, Eric Schain, Mariano Nastase, Samuel A. Zada, Zaid Grinstein-Dabus, Avigail Aubrey, Bobbi Gazula, Harshvardhan Feder, Amir Doyle, Werner Devore, Sasha Dugan, Patricia Friedman, Daniel Reichart, Roi Brenner, Michael Hassidim, Avinatan Devinsky, Orrin Flinker, Adeen Levy, Omer Hasson, Uri Department of Psychology The Neuroscience Institute Princeton University PrincetonNJ United States Google Research United States New York University Grossman School of Medicine New YorkNY United States School of Engineering and Applied Science Harvard University CambridgeMA United States New York University Tandon School of Engineering BrooklynNY United States Blavatnik School of Computer Science Tel-Aviv University Israel Department of Cognitive and Brain Sciences Business School Hebrew University of Jerusalem Israel Bioinformatics Interdepartmental Program University of California Los Angeles Los AngelesCA United States Technion - Israel Institute of Technology Haifa Israel

Deep Language Models (DLMs) provide a novel computational paradigm for understanding the mechanisms of natural language processing in the human brain. Unlike traditional psycholinguistic models, DLMs use layered sequences of continuous numerical vectors to represent words and context, allowing a plethora of emerging applications such as human-like text generation. In this paper we show evidence that the layered hierarchy of DLMs may be used to model the temporal dynamics of language comprehension in the brain by demonstrating a strong correlation between DLM layer depth and the time at which layers are most predictive of the human brain. Our ability to temporally resolve individual layers benefits from our use of electrocorticography (ECoG) data, which has a much higher temporal resolution than noninvasive methods like fMRI. Using ECoG, we record neural activity from participants listening to a 30-minute narrative while also feeding the same narrative to a high-performing DLM (GPT2-XL). We then extract contextual embeddings from the different layers of the DLM and use linear encoding models to predict neural activity. We first focus on the Inferior Frontal Gyrus (IFG, or Broca’s area) and then extend our model to track the increasing temporal receptive window along the linguistic processing hierarchy from auditory to syntactic and semantic areas. Our results reveal a connection between human language processing and DLMs, with the DLM’s layer-by-layer accumulation of contextual information mirroring the timing of neural activity in high-order language areas. © 2023, CC BY-NC-ND.

关键词： brain

How much competence is there in performance? Assessing the distributional hypothesis in word bigrams 6

学校读者我要写书评

暂无评论

How much competence is there in performance? Assessing the d...

6th Italian Conference on Computational Linguistics, CLiC-it 2019

作者： Seltmann, Johann Ducceschi, Luca Herbelot, Aurélie University of Potsdam Department of Linguistics Dept. of Information Engineering and Computer Science Germany University of Trento Dept. of Psychology and Cognitive Science Dept. of Information Engineering and Computer Science Italy University of Trento Center for Mind/Brain Sciences Dept. of Information Engineering and Computer Science Italy

The field of Distributional Semantics (DS) is built on the ‘distributional hypothesis’, which states that meaning can be recovered from statistical information in observable language. It is however notable that the computations necessary to obtain ‘good’ DS representations are often very involved, implying that if meaning is derivable from linguistic data, it is not directly encoded in it. This prompts questions related to fundamental questions about language acquisition: if we regard text data as linguistic performance, what kind of ‘innate’ mechanisms must operate over that data to reach competence? In other words, how much of semantic acquisition is truly data-driven, and what must be hard-encoded in a system’s architecture? In this paper, we introduce a new methodology to pull those questions apart. We use state-of-the-art computational models to investigate the amount and nature of transformations required to perform particular semantic tasks. We apply that methodology to one of the simplest structures in language: the word bigram, giving insights into the specific contribution of that linguistic component.1 Copyright © 2019 for this paper by its authors.

关键词： Semantics

Motor imagery enhances performance beyond the imagined action

学校读者我要写书评

暂无评论

Proceedings of the National Academy of sciences of the United States of America 2025年第20期122卷 e2423642122页

作者： Gippert, Magdalena Shih, Pei-Cheng Heed, Tobias Howard, Ian S. Jamshidi Idaji, Mina Villringer, Arno Sehm, Bernhard Nikulin, Vadim V. Department of Neurology Max Planck Institute for Human Cognitive and Brain Sciences Leipzig 04103 Germany Sony Computer Science Laboratories 141-0022 Tokyo Japan Department of Psychology and Centre for Cognitive Neuroscience University of Salzburg 5020 Salzburg Austria School of Engineering Computing and Mathematics Faculty of Science and Engineering University of Plymouth Plymouth PL4 8AA United Kingdom Machine Learning Group Berlin Institute for the Foundations of Learning and Data 10587 Berlin Germany Machine Learning Group Institute of Software Engineering and Theoretical Computer Science Electrical Engineering and Computer Science Faculty Technical University Berlin 10587 Berlin Germany Department of Neurology Martin Luther University of Halle-Wittenberg Halle (Saale) 06120 Germany

Motor imagery is frequently utilized to improve the performance of specific target movements in sports and rehabilitation. In this study, we show that motor imagery can facilitate learning of not only the imagined target movements but also sequentially linked overt movements. Hybrid sequences comprising imagined and physically executed segments allowed participants to learn specific movement characteristics of the executed segments when they were consistently associated with specific imagined segments. Electrophysiological recordings revealed that the degree of event-related synchronization in the alpha and beta bands during a basic motor imagery task was correlated with imagery-evoked motor learning. Thus, both behavioral and neural evidence indicate that motor imagery's benefits extend beyond the imagined movements, improving performance in linked overt movements. This provides decisive evidence for the functional equivalence of imagined and overt movements and suggests applications for imagery in sports and rehabilitation.

关键词： adaptation force field motor imagery motor learning motor sequence

DreamCoder: Growing generalizable, interpretable knowledge with wake-sleep Bayesian program learning

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Ellis, Kevin Wong, Catherine Nye, Maxwell Sablé-Meyer, Mathias Cary, Luc Morales, Lucas Hewitt, Luke Solar-Lezama, Armando Tenenbaum, Joshua B. MIT CSAIL NeuroSpin Center for Brains Minds and Machines Department of Brain and Cognitive Sciences Department of Electrical Engineering and Computer Science

Expert problem-solving is driven by powerful languages for thinking about problems and their solutions. Acquiring expertise means learning these languages — systems of concepts, alongside the skills to use them. We present DreamCoder, a system that learns to solve problems by writing programs. It builds expertise by creating programming languages for expressing domain concepts, together with neural networks to guide the search for programs within these languages. A "wake-sleep" learning algorithm alternately extends the language with new symbolic abstractions and trains the neural network on imagined and replayed problems. DreamCoder solves both classic inductive programming tasks and creative tasks such as drawing pictures and building scenes. It rediscovers the basics of modern functional programming, vector algebra and classical physics, including Newton’s and Coulomb’s laws. Concepts are built compositionally from those learned earlier, yielding multilayered symbolic representations that are interpretable and transferrable to new tasks, while still growing scalably and flexibly with experience. Copyright © 2020, The Authors. All rights reserved.

关键词： Wakes

GRADIENTS OF brain ORGANIZATION: SMOOTH SAILING FROM METHODS DEVELOPMENT TO USER COMMUNITY

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Royer, Jessica Paquola, Casey Valk, Sofie L. Kirschner, Matthias Hong, Seok-Jun Park, Bo-Yong Bethlehem, Richard A.I. Leech, Robert Yeo, B.T. Thomas Jefferies, Elizabeth Smallwood, Jonathan Margulies, Daniel Bernhardt, Boris C. Montreal Neurological Institute and Hospital McGill University MontrealQC Canada Forschungszentrum Julich Julich Germany Max Planck Institute for Human Cognitive and Brain Sciences Leipzig Germany Division of Adult Psychiatry Department of Psychiatry University Hospitals of Geneva Thonex Switzerland Center for Neuroscience Imaging Research Institute for Basic Science Suwon Korea Republic of Center for the Developing Brain Child Mind Institute New York United States Department of Biomedical Engineering Sungkyunkwan University Suwon Korea Republic of Department of Data Science Inha University Incheon Korea Republic of Department of Statistics and Data Science Inha University Incheon Korea Republic of Department of Psychology University of Cambridge Cambridge United Kingdom Centre for Neuroimaging Science King's College London London United Kingdom Centre for Sleep áCognition Centre for Translational Magnetic Resonance Research Yong Loo Lin School of Medicine National University of Singapore Singapore Department of Electrical and Computer Engineering National University of Singapore Singapore Singapore Institute for Health Institute for Digital Medicine National University of Singapore Singapore Singapore National University of Singapore Singapore Singapore Martinos Center for Biomedical Imaging Massachusetts General Hospital CharlestownMA United States Department of Psychology University of York York United Kingdom Department of Psychology Queens University KingstonON Canada Universite de Paris Paris France

Multimodal neuroimaging grants a powerful in vivo window into the structure and function of the human brain. Recent methodological and conceptual advances have enabled investigations of the interplay between large-scale spatial trends - or gradients - in brain structure and function, offering a framework to unify principles of brain organization across multiple scales. Strong community enthusiasm for these techniques has been instrumental in their widespread adoption and implementation to answer key questions in neuroscience. Following a brief review of current literature on this framework, this perspective paper will highlight how pragmatic steps aiming to make gradient methods more accessible to the community propelled these techniques to the forefront of neuroscientific inquiry. More specifically, we will emphasize how interest for gradient methods was catalyzed by data sharing, open-source software development, as well as the organization of dedicated workshops led by a diverse team of early career researchers. To this end, we argue that the growing excitement for brain gradients is the result of coordinated and consistent efforts to build an inclusive community and can serve as a case in point for future innovations and conceptual advances in neuroinformatics. We close this perspective paper by discussing challenges for the continuous refinement of neuroscientific theory, methodological innovation, and real-world translation to maintain our collective progress towards integrated models of brain organization. © 2024, CC BY.

关键词： Gradient methods

Comparison of Different Direct Deep Learning-based Attenuation Correction Schemes for Myocardial Perfusion SPECT

学校读者我要写书评

暂无评论

Comparison of Different Direct Deep Learning-based Attenuati...

IEEE Symposium on Nuclear science (NSS/MIC)

作者： Y. Du B. Wang C.-N. Lin G.-U. Hung G. S.P. Mok Department of Electrical and Computer Engineering Biomedical Imaging Laboratory (BIG) Taipa University of Macau Faculty of Science and Technology Macao Special Administrative Region China Institute of Collaborative Innovation Center for Cognitive and Brain Sciences Taipa University of Macau Macao Special Administrative Region China Chinese Academy of Sciences Shenzhen Institute of Advanced Technology Lauterbur Research Center for Biomedical Imaging Shenzhen China Department of Nuclear Medicine Show Chwan Memorial Hospital Changhua County Taiwan Department of Nuclear Medicine Chang Bing Show Chwan Memorial Hospital Changhua County Taiwan

ISBN: (数字)9798350388152

ISBN: (纸本)9798350388169

Deep learning-based attenuation correction (DLAC) has been proven to improve the quantitative and diagnostic accuracy of myocardial perfusion SPECT (MP SPECT) as compared to non-attenuation corrected (NAC) MP SPECT. Previous studies have investigated various DLAC strategies using different NAC input, e.g., volume-of-interests (VOI) of whole thoracic volume-of-interest (wVOI) for conventional scanners, cardiac region (cVOI) for dedicated cardiac scanner, and 2D polar plots (pROI). This study aims to provide a direct comparison of these DLAC methods. Three hundred clinical patients scanned with standard one-day stress/rest 99m Tc-sestamibi protocol are retrospectively recruited. 64 projections were acquired from RAO to LPO with 20 s/projection using a standard general-purpose dual-head SPECT scanner, and reconstructed with DEW scatter correction, with or without CTAC by 3D OS-EM algorithms with 48 updates. A conditional generative adversarial network was implemented to generated AC images from NAC images for 4 different inputs: 64×64×N wVOI, 24×24×24 cVOI, 24×24×24 short axis (SA) VOI, and 96×96 pROI. Both stress and rest scans were used, i.e., a total of 600 scans, dividing into 420, 60, and 120 for training, validation, and testing. After testing, all processed images were converted to polar plots for final comparison. Voxel-based bias and intraclass correlation analysis on extracted radiomics features were assessed using CTAC polar plots as reference. The training time for wVOI, cVOI, SA, and pROI was 2.5 hr, 1.0 hr, 1.0 hr, and 0.5 hr. We find that the inputs using all 3D VOIs have close quantitative and ICC results, while polar plot inputs are slightly worse. The use of wVOI input has the best ICC reliability with CTAC among all the VOIs. DLAC is robust for different types of NAC input.

关键词： Training Three-dimensional displays Semiconductor detectors Myocardium Feature extraction Attenuation Single photon emission computed tomography

A BCI based smart home system combined with event-related potentials and speech imagery task

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Kim, Hyeong-Jin Lee, Min-Ho Lee, Minji Department of Brain and Cognitive Engineering Korea University Seoul Korea Republic of Department of Computer Science Nazarbayev University Nur-Sultan Kazakhstan

Recently, smart home systems based on braincomputer interface (BCI) has attracted a wide range of interests in both industry and academia. However, the current BCI system has several shortcomings as it produces a comparatively lower accuracy for real-time implementations as well as the intuitive paradigm for the users cannot be well established here. Therefore, in this study, we proposed a highly intuitive BCI paradigm that combines event-related potential (ERP) with the speech-imagery task for the individual target objects. The decoding accuracy of the proposed paradigm was 88.1% ( ± 5.90) which is a much significant higher performance than a conventional ERP system. Furthermore, the amplitude of N700 components was significantly enhanced over frontal regions which are priory evoked by the speech-imagery task. Our results could be utilized to develop a smart home system so that it could be more user-friendly and convenient by means of delivering user's intentions both, intuitively and accurately. Copyright © 2020, The Authors. All rights reserved.

关键词： Automation

On cross-corpus generalization of deep learning based speech enhancement

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Pandey, Ashutosh Wang, DeLiang Department of Computer Science and Engineering Ohio State University ColumbusOH43210 United States Department of Computer Science and Engineering Center for Cognitive and Brain Sciences Ohio State University ColumbusOH43210 United States

In recent years, supervised approaches using deep neural networks (DNNs) have become the mainstream for speech enhancement. It has been established that DNNs generalize well to untrained noises and speakers if trained using a large number of noises and speakers. However, we find that DNNs fail to generalize to new speech corpora in low signal-to-noise ratio (SNR) conditions. In this work, we establish that the lack of generalization is mainly due to the channel mismatch between the trained and untrained corpus. Additionally, we observe that traditional channel normalization techniques are not effective in improving cross-corpus generalization. Further, we evaluate publicly available datasets that are promising for generalization. We find one particular corpus to be significantly better than others. Finally, we find that using a smaller frame shift in short-time processing of speech can significantly improve cross-corpus generalization. The proposed techniques to address cross-corpus generalization include channel normalization, better training corpus, and smaller frame shift in short-time Fourier transform (STFT). These techniques together improve the objective intelligibility and quality scores on untrained corpora significantly. Copyright © 2020, The Authors. All rights reserved.

关键词： Speech enhancement

An investigation on deep learning with Beta stabilizer

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Liu, Qi Tan, Tian Yu, Kai Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering SpeechLab Department of Computer Science and Engineering Brain Science and Technology Research Center Shanghai Jiao Tong University Shanghai China

Artificial neural networks (ANN) have been used in many applications such like handwriting recognition and speech recognition. It is well-known that learning rate is a crucial value in the training procedure for artificial neural networks. It is shown that the initial value of learning rate can confoundedly affect the final result and this value is always set manually in practice. A new parameter called beta stabilizer has been introduced to reduce the sensitivity of the initial learning rate. But this method has only been proposed for deep neural network (DNN) with sigmoid activation function. In this paper we extended beta stabilizer to long short-term memory (LSTM) and investigated the effects of beta stabilizer parameters on different models, including LSTM and DNN with relu activation function. It is concluded that beta stabilizer parameters can reduce the sensitivity of learning rate with almost the same performance on DNN with relu activation function and LSTM. However, it is shown that the effects of beta stabilizer on DNN with relu activation function and LSTM are fewer than the effects on DNN with sigmoid activation function. Copyright © 2020, The Authors. All rights reserved.

关键词： Long short-term memory