“Virtual” attenuation correction: improving stress myocardial perfusion SPECT imaging using deep learning DOI
Tomoe Hagio, Alexis Poitrasson-Rivière, Jonathan B. Moody

et al.

European Journal of Nuclear Medicine and Molecular Imaging, Journal Year: 2022, Volume and Issue: 49(9), P. 3140 - 3149

Published: March 21, 2022

Language: Английский

Convolutional Neural Networks as a Model of the Visual System: Past, Present, and Future DOI
Grace W. Lindsay

Journal of Cognitive Neuroscience, Journal Year: 2020, Volume and Issue: 33(10), P. 2017 - 2031

Published: Feb. 6, 2020

Abstract Convolutional neural networks (CNNs) were inspired by early findings in the study of biological vision. They have since become successful tools computer vision and state-of-the-art models both activity behavior on visual tasks. This review highlights what, context CNNs, it means to be a good model computational neuroscience various ways can provide insight. Specifically, covers origins CNNs methods which we validate them as It then goes elaborate what learn about understanding experimenting discusses emerging opportunities for use research beyond basic object recognition.

Language: Английский

Citations

446

Deep learning tools for the measurement of animal behavior in neuroscience DOI
Mackenzie Weygandt Mathis, Alexander Mathis

Current Opinion in Neurobiology, Journal Year: 2019, Volume and Issue: 60, P. 1 - 11

Published: Nov. 29, 2019

Language: Английский

Citations

382

The neuroconnectionist research programme DOI
Adrien Doerig,

Rowan P. Sommers,

Katja Seeliger

et al.

Nature reviews. Neuroscience, Journal Year: 2023, Volume and Issue: 24(7), P. 431 - 450

Published: May 30, 2023

Language: Английский

Citations

134

Applications of deep learning in precision weed management: A review DOI Creative Commons
Nitin Rai, Yu Zhang, Billy G. Ram

et al.

Computers and Electronics in Agriculture, Journal Year: 2023, Volume and Issue: 206, P. 107698 - 107698

Published: Feb. 10, 2023

Deep Learning (DL) has been described as one of the key subfields Artificial Intelligence (AI) that is transforming weed detection for site-specific management (SSWM). In last demi-decade, DL techniques have integrated with ground well aerial-based technologies to identify weeds in still image context and real-time setting. After observing current research trend DL-based detection, are advancing by assisting precision weeding make smart decisions. Therefore, objective this paper was present a systematic review study involves available SSWM. To accomplish study, comprehensive literature survey performed consists 60 closest technical papers on detection. The findings summarized follows, a) transfer learning approach widely adopted technique address majority work, b) less focus navigated towards custom designed neural networks task, c) based pretrained models deployed test dataset, no specific model can be attributed achieved high accuracy multiple field images pertaining several studies, d) inferencing resource-constrained edge devices limited number dataset lagging, e) different versions YOLO (mostly v3) detecting scenario, f) SegNet U-Net semantic segmentation task multispectral aerial imagery, g) open-source acquired using drones, h) lack exploring optimization generalization identification images, i) ways design consume training hours, low-power consumption parameters during or inferencing, j) slow-moving advances optimizing domain adaptation approach. conclusion, will help researchers, experts, scientists, farmers, technology extension specialist gain updates area

Language: Английский

Citations

116

Limits to visual representational correspondence between convolutional neural networks and the human brain DOI Creative Commons
Yaoda Xu, Maryam Vaziri-Pashkam

Nature Communications, Journal Year: 2021, Volume and Issue: 12(1)

Published: April 6, 2021

Abstract Convolutional neural networks (CNNs) are increasingly used to model human vision due their high object categorization capabilities and general correspondence with brain responses. Here we evaluate the performance of 14 different CNNs compared fMRI responses natural artificial images using representational similarity analysis. Despite presence some CNN-brain CNNs’ impressive ability fully capture lower level visual representation real-world objects, show that do not higher representations nor those either at or levels representations. The latter is particularly critical, as processing both stimuli engages same circuits. We report similar results regardless differences in CNN architecture, training, recurrent processing. This indicates fundamental exist how represent information.

Language: Английский

Citations

115

Improving the accuracy of single-trial fMRI response estimates using GLMsingle DOI Creative Commons
Jacob S. Prince, Ian Charest, Jan W. Kurzawski

et al.

eLife, Journal Year: 2022, Volume and Issue: 11

Published: Nov. 29, 2022

Advances in artificial intelligence have inspired a paradigm shift human neuroscience, yielding large-scale functional magnetic resonance imaging (fMRI) datasets that provide high-resolution brain responses to thousands of naturalistic visual stimuli. Because such experiments necessarily involve brief stimulus durations and few repetitions each stimulus, achieving sufficient signal-to-noise ratio can be major challenge. We address this challenge by introducing GLMsingle , scalable, user-friendly toolbox available MATLAB Python enables accurate estimation single-trial fMRI ( glmsingle.org ). Requiring only time-series data design matrix as inputs, integrates three techniques for improving the accuracy trial-wise general linear model (GLM) beta estimates. First, voxel, custom hemodynamic response function (HRF) is identified from library candidate functions. Second, cross-validation used derive set noise regressors voxels unrelated experiment. Third, improve stability estimates closely spaced trials, betas are regularized on voxel-wise basis using ridge regression. Applying Natural Scenes Dataset BOLD5000, we find substantially improves reliability across visually-responsive cortex all subjects. Comparable improvements also observed smaller-scale auditory dataset StudyForrest These translate into tangible benefits higher-level analyses relevant systems cognitive neuroscience. demonstrate GLMsingle: (i) helps decorrelate between trials nearby time; (ii) enhances representational similarity subjects within datasets; (iii) boosts one-versus-many decoding publicly tool significantly quality past, present, future neuroimaging sampling activity many experimental conditions.

Language: Английский

Citations

92

A self-supervised domain-general learning framework for human ventral stream representation DOI Creative Commons
Talia Konkle, George A. Alvarez

Nature Communications, Journal Year: 2022, Volume and Issue: 13(1)

Published: Jan. 25, 2022

Abstract Anterior regions of the ventral visual stream encode substantial information about object categories. Are top-down category-level forces critical for arriving at this representation, or can representation be formed purely through domain-general learning natural image structure? Here we present a fully self-supervised model which learns to represent individual images, rather than categories, such that views same are embedded nearby in low-dimensional feature space, distinctly from other recently encountered views. We find category implicitly emerges local similarity structure space. Further, these models learn hierarchical features capture brain responses across human stream, on par with category-supervised models. These results provide computational support framework guiding formation where proximate goal is not explicitly information, but instead unique, compressed descriptions world.

Language: Английский

Citations

84

Physics-informed machine learning: A comprehensive review on applications in anomaly detection and condition monitoring DOI Creative Commons
Yuandi Wu, Brett Sicard, S. Andrew Gadsden

et al.

Expert Systems with Applications, Journal Year: 2024, Volume and Issue: 255, P. 124678 - 124678

Published: July 5, 2024

Condition monitoring plays a vital role in ensuring the reliability and optimal performance of various engineering systems. Traditional methods for condition rely on physics-based models statistical analysis techniques. However, these approaches often face challenges dealing with complex systems limited availability accurate physical models. In recent years, physics-informed machine learning (PIML) has emerged as promising approach monitoring, combining strengths modelling data-driven learning. This study presents comprehensive overview PIML techniques context monitoring. The central concept driving is incorporation known laws constraints into algorithms, enabling them to learn from available data while remaining consistent principles. Through fusing domain knowledge learning, offer enhanced accuracy interpretability comparison purely approaches. this survey, detailed examinations are performed regard methodology by which principles integrated within frameworks, well their suitability specific tasks Incorporation ML model may be realized variety methods, each having its unique advantages drawbacks. distinct limitations integration physics detailed, considering factors such computational efficiency, interpretability, generalizability different fault detection. Several case studies works literature utilizing emerging presented demonstrate efficacy applications. From reviewed, versatility potential demonstrated. Novel an innovative solution addressing complexities associated challenges. survey helps form foundation future work field. As technology continues advance, expected play crucial enhancing maintenance strategies, system reliability, overall operational efficiency

Language: Английский

Citations

29

Performance vs. competence in human–machine comparisons DOI Creative Commons
Chaz Firestone

Proceedings of the National Academy of Sciences, Journal Year: 2020, Volume and Issue: 117(43), P. 26562 - 26571

Published: Oct. 13, 2020

Does the human mind resemble machines that can behave like it? Biologically inspired machine-learning systems approach “human-level” accuracy in an astounding variety of domains, and even predict brain activity—raising exciting possibility such represent world we do. However, seemingly intelligent fail strange “unhumanlike” ways, threatening their status as models our minds. How know when human–machine behavioral differences reflect deep disparities underlying capacities, vs. failures are only superficial or peripheral? This article draws on a foundational insight from cognitive science—the distinction between performance competence —to encourage “species-fair” comparisons humans machines. The performance/competence urges us to consider whether failure system ideally hypothesized, one creature another, arises not because lacks relevant knowledge internal capacities (“competence”), but instead constraints demonstrating (“performance”). I argue this has been neglected by research comparing machine behavior, it should be essential any comparison. Focusing domain image classification, identify three factors contributing species-fairness comparisons, extracted recent work equates constraints. Species-fair level playing field natural artificial intelligence, so separate more those may enduring.

Language: Английский

Citations

125

A bird’s-eye view of deep learning in bioimage analysis DOI Creative Commons
Erik Meijering

Computational and Structural Biotechnology Journal, Journal Year: 2020, Volume and Issue: 18, P. 2312 - 2325

Published: Jan. 1, 2020

Deep learning of artificial neural networks has become the de facto standard approach to solving data analysis problems in virtually all fields science and engineering. Also biology medicine, deep technologies are fundamentally transforming how we acquire, process, analyze, interpret data, with potentially far-reaching consequences for healthcare. In this mini-review, take a bird’s-eye view at past, present, future developments learning, starting from large, biomedical imaging, bioimage particular.

Language: Английский

Citations

118