Xuebin Wang, Chunxiuzi Liu, Meng Zhao, Ke Zhang, Zengru Di, He Liu
This study introduces an artificial neural network (ANN) for image classification task, inspired by the aversive olfactory learning circuits of the nematode Caenorhabditis elegans (C. elegans). Despite the remarkable performance of ANNs in a variety of tasks, they face challenges such as excessive parameterization, high training costs and limited generalization capabilities. C. elegans, with its simple nervous system comprising only 302 neurons, serves as a paradigm in neurobiological research and is capable of complex behaviors including learning. This research identifies key neural circuits associated with aversive olfactory learning in C. elegans through behavioral experiments and high-throughput gene sequencing, translating them into an image classification ANN architecture. Additionally, two other image classification ANNs with distinct architectures were constructed for comparative performance analysis to highlight the advantages of bio-inspired design. The results indicate that the ANN inspired by the aversive olfactory learning circuits of C. elegans achieves higher accuracy, better consistency and faster convergence rates in image classification task, especially when tackling more complex classification challenges. This study not only showcases the potential of bio-inspired design in enhancing ANN capabilities but also provides a novel perspective and methodology for future ANN design.
本研究介绍了一种用于图像分类任务的人工神经网络（ANN），其灵感来源于线虫高脚线虫（C. elegans）的厌恶嗅觉学习回路。尽管人工神经网络在各种任务中表现出色，但它们也面临着参数过多、训练成本高和泛化能力有限等挑战。草履虫的神经系统非常简单，只有 302 个神经元，但它是神经生物学研究的典范，能够进行包括学习在内的复杂行为。这项研究通过行为实验和高通量基因测序，确定了与线虫厌恶性嗅觉学习相关的关键神经回路，并将其转化为图像分类 ANN 架构。此外，为了突出生物启发设计的优势，研究人员还构建了另外两个具有不同架构的图像分类 ANN 进行性能对比分析。研究结果表明，受优雅蛇的厌恶嗅觉学习电路启发的自动分类网络在图像分类任务中实现了更高的准确率、更好的一致性和更快的收敛速度，尤其是在应对更复杂的分类挑战时。这项研究不仅展示了生物启发设计在提高自动识别网络能力方面的潜力，还为未来的自动识别网络设计提供了新的视角和方法。
{"title":"An Artificial Neural Network for Image Classification Inspired by Aversive Olfactory Learning Circuits in Caenorhabditis Elegans","authors":"Xuebin Wang, Chunxiuzi Liu, Meng Zhao, Ke Zhang, Zengru Di, He Liu","doi":"arxiv-2409.07466","DOIUrl":"https://doi.org/arxiv-2409.07466","url":null,"abstract":"This study introduces an artificial neural network (ANN) for image\u0000classification task, inspired by the aversive olfactory learning circuits of\u0000the nematode Caenorhabditis elegans (C. elegans). Despite the remarkable\u0000performance of ANNs in a variety of tasks, they face challenges such as\u0000excessive parameterization, high training costs and limited generalization\u0000capabilities. C. elegans, with its simple nervous system comprising only 302\u0000neurons, serves as a paradigm in neurobiological research and is capable of\u0000complex behaviors including learning. This research identifies key neural\u0000circuits associated with aversive olfactory learning in C. elegans through\u0000behavioral experiments and high-throughput gene sequencing, translating them\u0000into an image classification ANN architecture. Additionally, two other image\u0000classification ANNs with distinct architectures were constructed for\u0000comparative performance analysis to highlight the advantages of bio-inspired\u0000design. The results indicate that the ANN inspired by the aversive olfactory\u0000learning circuits of C. elegans achieves higher accuracy, better consistency\u0000and faster convergence rates in image classification task, especially when\u0000tackling more complex classification challenges. This study not only showcases\u0000the potential of bio-inspired design in enhancing ANN capabilities but also\u0000provides a novel perspective and methodology for future ANN design.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211864","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Mara Barberis, Pieter De Clercq, Bastiaan Tamm, Hugo Van hamme, Maaike Vandermosten
Aphasia is a language disorder affecting one third of stroke patients. Current aphasia assessment does not consider natural speech due to the time consuming nature of manual transcriptions and a lack of knowledge on how to analyze such data. Here, we evaluate the potential of automatic speech recognition (ASR) to transcribe Dutch aphasic speech and the ability of natural speech features to detect aphasia. A picture-description task was administered and automatically transcribed in 62 persons with aphasia and 57 controls. Acoustic and linguistic features were semi-automatically extracted and provided as input to a support vector machine (SVM) classifier. Our ASR model obtained a WER of 24.5%, outperforming earlier ASR models for aphasia. The SVM shows high accuracy (86.6%) at the individual level, with fluency features as most dominant to detect aphasia. ASR and semi-automatic feature extraction can thus facilitate natural speech analysis in a time efficient manner in clinical practice.
失语症是一种语言障碍，影响着三分之一的中风患者。目前的失语症评估并不考虑自然语音，原因是人工转录耗时，而且缺乏如何分析此类数据的知识。在此，我们评估了自动语音识别（ASR）转录荷兰语失语语音的潜力以及自然语音特征检测失语症的能力。我们对 62 名失语症患者和 57 名对照组患者的图片描述任务进行了自动转录，并半自动提取了声学和语言特征，作为支持向量机 (SVM) 分类器的输入。我们的 ASR 模型的误码率为 24.5%，优于早期的失语症 ASR 模型。SVM 在个体水平上显示出较高的准确率（86.6%），其中流利度特征是检测失语症的最主要特征。因此，在临床实践中，ASR 和半自动特征提取能够以省时省力的方式促进自然语音分析。
{"title":"Automatic recognition and detection of aphasic natural speech","authors":"Mara Barberis, Pieter De Clercq, Bastiaan Tamm, Hugo Van hamme, Maaike Vandermosten","doi":"arxiv-2408.14082","DOIUrl":"https://doi.org/arxiv-2408.14082","url":null,"abstract":"Aphasia is a language disorder affecting one third of stroke patients.\u0000Current aphasia assessment does not consider natural speech due to the time\u0000consuming nature of manual transcriptions and a lack of knowledge on how to\u0000analyze such data. Here, we evaluate the potential of automatic speech\u0000recognition (ASR) to transcribe Dutch aphasic speech and the ability of natural\u0000speech features to detect aphasia. A picture-description task was administered\u0000and automatically transcribed in 62 persons with aphasia and 57 controls.\u0000Acoustic and linguistic features were semi-automatically extracted and provided\u0000as input to a support vector machine (SVM) classifier. Our ASR model obtained a\u0000WER of 24.5%, outperforming earlier ASR models for aphasia. The SVM shows high\u0000accuracy (86.6%) at the individual level, with fluency features as most\u0000dominant to detect aphasia. ASR and semi-automatic feature extraction can thus\u0000facilitate natural speech analysis in a time efficient manner in clinical\u0000practice.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211866","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A fundamental paradigm in neuroscience is that cognitive functions -- such as perception, learning, memory, and locomotion -- are governed by the brain's structural organization. Yet, the theoretical principles explaining how the physical architecture of the nervous system shapes its function remain elusive. Here, we combine concepts from quantum statistical mechanics and graph C*-algebras to introduce a theoretical framework where functional states of a structural connectome emerge as thermal equilibrium states of the underlying directed network. These equilibrium states, defined from the Kubo-Martin-Schwinger states formalism (KMS states), quantify the relative contribution of each neuron to the information flow within the connectome. Using the prototypical connectome of the nematode {em Caenorhabditis elegans}, we provide a comprehensive description of these KMS states, explore their functional implications, and establish the predicted functional network based on the nervous system's anatomical connectivity. Ultimately, we present a model for identifying the potential functional states of a detailed structural connectome and for conceptualizing the structure-function relationship.
神经科学的一个基本范式是认知功能（如感知、学习、记忆和运动）受大脑结构组织的支配。在这里，我们结合量子统计力学和图 C* 矩阵的概念，引入了一个理论框架，在这个框架中，结构连接体的功能状态是作为底层定向网络的热平衡状态出现的。这些平衡态由库勃-马丁-施文格状态形式主义（Kubo-Martin-Schwinger states formalism，KMS状态）定义，量化了每个神经元对连接组内信息流的相对贡献。我们利用线虫{（Caemenorhabditis elegans}}的原型连接组，全面描述了这些KMS状态，探讨了它们的功能含义，并根据神经系统的解剖连接建立了预测的功能网络。最终，我们提出了一种模式，用于识别详细结构连接组的潜在功能状态，并将结构与功能的关系概念化。
{"title":"Emergence of brain function from structure: an algebraic quantum model","authors":"Elkaïoum M. Moutuou, Habib Benali","doi":"arxiv-2408.14221","DOIUrl":"https://doi.org/arxiv-2408.14221","url":null,"abstract":"A fundamental paradigm in neuroscience is that cognitive functions -- such as\u0000perception, learning, memory, and locomotion -- are governed by the brain's\u0000structural organization. Yet, the theoretical principles explaining how the\u0000physical architecture of the nervous system shapes its function remain elusive.\u0000Here, we combine concepts from quantum statistical mechanics and graph\u0000C*-algebras to introduce a theoretical framework where functional states of a\u0000structural connectome emerge as thermal equilibrium states of the underlying\u0000directed network. These equilibrium states, defined from the\u0000Kubo-Martin-Schwinger states formalism (KMS states), quantify the relative\u0000contribution of each neuron to the information flow within the connectome.\u0000Using the prototypical connectome of the nematode {em Caenorhabditis elegans},\u0000we provide a comprehensive description of these KMS states, explore their\u0000functional implications, and establish the predicted functional network based\u0000on the nervous system's anatomical connectivity. Ultimately, we present a model\u0000for identifying the potential functional states of a detailed structural\u0000connectome and for conceptualizing the structure-function relationship.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211865","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Gang Qu, Ziyu Zhou, Vince D. Calhoun, Aiying Zhang, Yu-Ping Wang
Multimodal neuroimaging modeling has becomes a widely used approach but confronts considerable challenges due to heterogeneity, which encompasses variability in data types, scales, and formats across modalities. This variability necessitates the deployment of advanced computational methods to integrate and interpret these diverse datasets within a cohesive analytical framework. In our research, we amalgamate functional magnetic resonance imaging, diffusion tensor imaging, and structural MRI into a cohesive framework. This integration capitalizes on the unique strengths of each modality and their inherent interconnections, aiming for a comprehensive understanding of the brain's connectivity and anatomical characteristics. Utilizing the Glasser atlas for parcellation, we integrate imaging derived features from various modalities: functional connectivity from fMRI, structural connectivity from DTI, and anatomical features from sMRI within consistent regions. Our approach incorporates a masking strategy to differentially weight neural connections, thereby facilitating a holistic amalgamation of multimodal imaging data. This technique enhances interpretability at connectivity level, transcending traditional analyses centered on singular regional attributes. The model is applied to the Human Connectome Project's Development study to elucidate the associations between multimodal imaging and cognitive functions throughout youth. The analysis demonstrates improved predictive accuracy and uncovers crucial anatomical features and essential neural connections, deepening our understanding of brain structure and function.
{"title":"Integrated Brain Connectivity Analysis with fMRI, DTI, and sMRI Powered by Interpretable Graph Neural Networks","authors":"Gang Qu, Ziyu Zhou, Vince D. Calhoun, Aiying Zhang, Yu-Ping Wang","doi":"arxiv-2408.14254","DOIUrl":"https://doi.org/arxiv-2408.14254","url":null,"abstract":"Multimodal neuroimaging modeling has becomes a widely used approach but\u0000confronts considerable challenges due to heterogeneity, which encompasses\u0000variability in data types, scales, and formats across modalities. This\u0000variability necessitates the deployment of advanced computational methods to\u0000integrate and interpret these diverse datasets within a cohesive analytical\u0000framework. In our research, we amalgamate functional magnetic resonance\u0000imaging, diffusion tensor imaging, and structural MRI into a cohesive\u0000framework. This integration capitalizes on the unique strengths of each\u0000modality and their inherent interconnections, aiming for a comprehensive\u0000understanding of the brain's connectivity and anatomical characteristics.\u0000Utilizing the Glasser atlas for parcellation, we integrate imaging derived\u0000features from various modalities: functional connectivity from fMRI, structural\u0000connectivity from DTI, and anatomical features from sMRI within consistent\u0000regions. Our approach incorporates a masking strategy to differentially weight\u0000neural connections, thereby facilitating a holistic amalgamation of multimodal\u0000imaging data. This technique enhances interpretability at connectivity level,\u0000transcending traditional analyses centered on singular regional attributes. The\u0000model is applied to the Human Connectome Project's Development study to\u0000elucidate the associations between multimodal imaging and cognitive functions\u0000throughout youth. The analysis demonstrates improved predictive accuracy and\u0000uncovers crucial anatomical features and essential neural connections,\u0000deepening our understanding of brain structure and function.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211868","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, we propose a novel framework for multimodal contrastive learning utilizing a quantum encoder to integrate EEG (electroencephalogram) and image data. This groundbreaking attempt explores the integration of quantum encoders within the traditional multimodal learning framework. By leveraging the unique properties of quantum computing, our method enhances the representation learning capabilities, providing a robust framework for analyzing time series and visual information concurrently. We demonstrate that the quantum encoder effectively captures intricate patterns within EEG signals and image features, facilitating improved contrastive learning across modalities. This work opens new avenues for integrating quantum computing with multimodal data analysis, particularly in applications requiring simultaneous interpretation of temporal and visual data.
{"title":"Quantum Multimodal Contrastive Learning Framework","authors":"Chi-Sheng Chen, Aidan Hung-Wen Tsai, Sheng-Chieh Huang","doi":"arxiv-2408.13919","DOIUrl":"https://doi.org/arxiv-2408.13919","url":null,"abstract":"In this paper, we propose a novel framework for multimodal contrastive\u0000learning utilizing a quantum encoder to integrate EEG (electroencephalogram)\u0000and image data. This groundbreaking attempt explores the integration of quantum\u0000encoders within the traditional multimodal learning framework. By leveraging\u0000the unique properties of quantum computing, our method enhances the\u0000representation learning capabilities, providing a robust framework for\u0000analyzing time series and visual information concurrently. We demonstrate that\u0000the quantum encoder effectively captures intricate patterns within EEG signals\u0000and image features, facilitating improved contrastive learning across\u0000modalities. This work opens new avenues for integrating quantum computing with\u0000multimodal data analysis, particularly in applications requiring simultaneous\u0000interpretation of temporal and visual data.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211869","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A suite of impressive scientific discoveries have been driven by recent advances in artificial intelligence. These almost all result from training flexible algorithms to solve difficult optimization problems specified in advance by teams of domain scientists and engineers with access to large amounts of data. Although extremely useful, this kind of problem solving only corresponds to one part of science - the "easy problem." The other part of scientific research is coming up with the problem itself - the "hard problem." Solving the hard problem is beyond the capacities of current algorithms for scientific discovery because it requires continual conceptual revision based on poorly defined constraints. We can make progress on understanding how humans solve the hard problem by studying the cognitive science of scientists, and then use the results to design new computational agents that automatically infer and update their scientific paradigms.
{"title":"Artificial intelligence for science: The easy and hard problems","authors":"Ruairidh M. Battleday, Samuel J. Gershman","doi":"arxiv-2408.14508","DOIUrl":"https://doi.org/arxiv-2408.14508","url":null,"abstract":"A suite of impressive scientific discoveries have been driven by recent\u0000advances in artificial intelligence. These almost all result from training\u0000flexible algorithms to solve difficult optimization problems specified in\u0000advance by teams of domain scientists and engineers with access to large\u0000amounts of data. Although extremely useful, this kind of problem solving only\u0000corresponds to one part of science - the \"easy problem.\" The other part of\u0000scientific research is coming up with the problem itself - the \"hard problem.\"\u0000Solving the hard problem is beyond the capacities of current algorithms for\u0000scientific discovery because it requires continual conceptual revision based on\u0000poorly defined constraints. We can make progress on understanding how humans\u0000solve the hard problem by studying the cognitive science of scientists, and\u0000then use the results to design new computational agents that automatically\u0000infer and update their scientific paradigms.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142226738","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Italo Ivo Lima Dias Pinto, Zhibin Zhou, Javier O. Garcia, Ramesh Srinivasan
We propose a novel approach to investigate the brain mechanisms that support coordination of behavior between individuals. Brain states in single individuals defined by the patterns of functional connectivity between brain regions are used to create joint symbolic representations of the evolution of brain states in two or more individuals performing a task together. These symbolic dynamics can be analyzed to reveal aspects of the dynamics of joint brain states that are related to coordination or other interactive behaviors. We apply this approach to simultaneous electroencephalographic (EEG) data from pairs of subjects engaged in two different modes of finger-tapping coordination tasks (synchronization and syncopation) under different interaction conditions (Uncoupled, Leader-Follower, and Mutual) to explore the neural mechanisms of multi-person motor coordination. Our results reveal that the dyads exhibit mostly the same joint symbols in different interaction conditions - the most important differences are reflected in the symbolic dynamics. Recurrence analysis shows that interaction influences the dwell time in specific joint symbols and the structure of joint symbol sequences (motif length). In synchronization, increasing feedback promotes stability with longer dwell times and motif length. In syncopation, Leader-Follower interactions enhance stability (increase dwell time and motif length), but Mutual feedback dramatically reduces stability. Network analysis reveals distinct topological changes with task and feedback. In synchronization, stronger coupling stabilizes a few states restricting the pattern of flow between states, preserving a core-periphery structure of the joint brain states. In syncopation, a more distributed flow amongst a larger set of joint brain states reduces the dominance of core joint brain states.
{"title":"Symbolic dynamics of joint brain states during dyadic coordination","authors":"Italo Ivo Lima Dias Pinto, Zhibin Zhou, Javier O. Garcia, Ramesh Srinivasan","doi":"arxiv-2408.13360","DOIUrl":"https://doi.org/arxiv-2408.13360","url":null,"abstract":"We propose a novel approach to investigate the brain mechanisms that support\u0000coordination of behavior between individuals. Brain states in single\u0000individuals defined by the patterns of functional connectivity between brain\u0000regions are used to create joint symbolic representations of the evolution of\u0000brain states in two or more individuals performing a task together. These\u0000symbolic dynamics can be analyzed to reveal aspects of the dynamics of joint\u0000brain states that are related to coordination or other interactive behaviors.\u0000We apply this approach to simultaneous electroencephalographic (EEG) data from\u0000pairs of subjects engaged in two different modes of finger-tapping coordination\u0000tasks (synchronization and syncopation) under different interaction conditions\u0000(Uncoupled, Leader-Follower, and Mutual) to explore the neural mechanisms of\u0000multi-person motor coordination. Our results reveal that the dyads exhibit\u0000mostly the same joint symbols in different interaction conditions - the most\u0000important differences are reflected in the symbolic dynamics. Recurrence\u0000analysis shows that interaction influences the dwell time in specific joint\u0000symbols and the structure of joint symbol sequences (motif length). In\u0000synchronization, increasing feedback promotes stability with longer dwell times\u0000and motif length. In syncopation, Leader-Follower interactions enhance\u0000stability (increase dwell time and motif length), but Mutual feedback\u0000dramatically reduces stability. Network analysis reveals distinct topological\u0000changes with task and feedback. In synchronization, stronger coupling\u0000stabilizes a few states restricting the pattern of flow between states,\u0000preserving a core-periphery structure of the joint brain states. In\u0000syncopation, a more distributed flow amongst a larger set of joint brain states\u0000reduces the dominance of core joint brain states.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211867","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Do neural network models of vision learn brain-aligned representations because they share architectural constraints and task objectives with biological vision or because they learn universal features of natural image processing? We characterized the universality of hundreds of thousands of representational dimensions from visual neural networks with varied construction. We found that networks with varied architectures and task objectives learn to represent natural images using a shared set of latent dimensions, despite appearing highly distinct at a surface level. Next, by comparing these networks with human brain representations measured with fMRI, we found that the most brain-aligned representations in neural networks are those that are universal and independent of a network's specific characteristics. Remarkably, each network can be reduced to fewer than ten of its most universal dimensions with little impact on its representational similarity to the human brain. These results suggest that the underlying similarities between artificial and biological vision are primarily governed by a core set of universal image representations that are convergently learned by diverse systems.
{"title":"Universal dimensions of visual representation","authors":"Zirui Chen, Michael F. Bonner","doi":"arxiv-2408.12804","DOIUrl":"https://doi.org/arxiv-2408.12804","url":null,"abstract":"Do neural network models of vision learn brain-aligned representations\u0000because they share architectural constraints and task objectives with\u0000biological vision or because they learn universal features of natural image\u0000processing? We characterized the universality of hundreds of thousands of\u0000representational dimensions from visual neural networks with varied\u0000construction. We found that networks with varied architectures and task\u0000objectives learn to represent natural images using a shared set of latent\u0000dimensions, despite appearing highly distinct at a surface level. Next, by\u0000comparing these networks with human brain representations measured with fMRI,\u0000we found that the most brain-aligned representations in neural networks are\u0000those that are universal and independent of a network's specific\u0000characteristics. Remarkably, each network can be reduced to fewer than ten of\u0000its most universal dimensions with little impact on its representational\u0000similarity to the human brain. These results suggest that the underlying\u0000similarities between artificial and biological vision are primarily governed by\u0000a core set of universal image representations that are convergently learned by\u0000diverse systems.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211871","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We study pattern formation in class of a large-dimensional neural networks posed on random graphs and subject to spatio-temporal stochastic forcing. Under generic conditions on coupling and nodal dynamics, we prove that the network admits a rigorous mean-field limit, resembling a Wilson-Cowan neural field equation. The state variables of the limiting systems are the mean and variance of neuronal activity. We select networks whose mean-field equations are tractable and we perform a bifurcation analysis using as control parameter the diffusivity strength of the afferent white noise on each neuron. We find conditions for Turing-like bifurcations in a system where the cortex is modelled as a ring, and we produce numerical evidence of noise-induced spiral waves in models with a two-dimensional cortex. We provide numerical evidence that solutions of the finite-size network converge weakly to solutions of the mean-field model. Finally, we prove a Large Deviation Principle, which provides a means of assessing the likelihood of deviations from the mean-field equations induced by finite-size effects.
{"title":"Neural Fields and Noise-Induced Patterns in Neurons on Large Disordered Networks","authors":"Daniele Avitabile, James MacLaurin","doi":"arxiv-2408.12540","DOIUrl":"https://doi.org/arxiv-2408.12540","url":null,"abstract":"We study pattern formation in class of a large-dimensional neural networks\u0000posed on random graphs and subject to spatio-temporal stochastic forcing. Under\u0000generic conditions on coupling and nodal dynamics, we prove that the network\u0000admits a rigorous mean-field limit, resembling a Wilson-Cowan neural field\u0000equation. The state variables of the limiting systems are the mean and variance\u0000of neuronal activity. We select networks whose mean-field equations are\u0000tractable and we perform a bifurcation analysis using as control parameter the\u0000diffusivity strength of the afferent white noise on each neuron. We find\u0000conditions for Turing-like bifurcations in a system where the cortex is\u0000modelled as a ring, and we produce numerical evidence of noise-induced spiral\u0000waves in models with a two-dimensional cortex. We provide numerical evidence\u0000that solutions of the finite-size network converge weakly to solutions of the\u0000mean-field model. Finally, we prove a Large Deviation Principle, which provides\u0000a means of assessing the likelihood of deviations from the mean-field equations\u0000induced by finite-size effects.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211887","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Zhonghao He, Jascha Achterberg, Katie Collins, Kevin Nejad, Danyal Akarca, Yinzhu Yang, Wes Gurnee, Ilia Sucholutsky, Yuhan Tang, Rebeca Ianov, George Ogden, Chole Li, Kai Sandbrink, Stephen Casper, Anna Ivanova, Grace W. Lindsay
As deep learning systems are scaled up to many billions of parameters, relating their internal structure to external behaviors becomes very challenging. Although daunting, this problem is not new: Neuroscientists and cognitive scientists have accumulated decades of experience analyzing a particularly complex system - the brain. In this work, we argue that interpreting both biological and artificial neural systems requires analyzing those systems at multiple levels of analysis, with different analytic tools for each level. We first lay out a joint grand challenge among scientists who study the brain and who study artificial neural networks: understanding how distributed neural mechanisms give rise to complex cognition and behavior. We then present a series of analytical tools that can be used to analyze biological and artificial neural systems, organizing those tools according to Marr's three levels of analysis: computation/behavior, algorithm/representation, and implementation. Overall, the multilevel interpretability framework provides a principled way to tackle neural system complexity; links structure, computation, and behavior; clarifies assumptions and research priorities at each level; and paves the way toward a unified effort for understanding intelligent systems, may they be biological or artificial.
{"title":"Multilevel Interpretability Of Artificial Neural Networks: Leveraging Framework And Methods From Neuroscience","authors":"Zhonghao He, Jascha Achterberg, Katie Collins, Kevin Nejad, Danyal Akarca, Yinzhu Yang, Wes Gurnee, Ilia Sucholutsky, Yuhan Tang, Rebeca Ianov, George Ogden, Chole Li, Kai Sandbrink, Stephen Casper, Anna Ivanova, Grace W. Lindsay","doi":"arxiv-2408.12664","DOIUrl":"https://doi.org/arxiv-2408.12664","url":null,"abstract":"As deep learning systems are scaled up to many billions of parameters,\u0000relating their internal structure to external behaviors becomes very\u0000challenging. Although daunting, this problem is not new: Neuroscientists and\u0000cognitive scientists have accumulated decades of experience analyzing a\u0000particularly complex system - the brain. In this work, we argue that\u0000interpreting both biological and artificial neural systems requires analyzing\u0000those systems at multiple levels of analysis, with different analytic tools for\u0000each level. We first lay out a joint grand challenge among scientists who study\u0000the brain and who study artificial neural networks: understanding how\u0000distributed neural mechanisms give rise to complex cognition and behavior. We\u0000then present a series of analytical tools that can be used to analyze\u0000biological and artificial neural systems, organizing those tools according to\u0000Marr's three levels of analysis: computation/behavior,\u0000algorithm/representation, and implementation. Overall, the multilevel\u0000interpretability framework provides a principled way to tackle neural system\u0000complexity; links structure, computation, and behavior; clarifies assumptions\u0000and research priorities at each level; and paves the way toward a unified\u0000effort for understanding intelligent systems, may they be biological or\u0000artificial.","PeriodicalId":501517,"journal":{"name":"arXiv - QuanBio - Neurons and Cognition","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142211870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：