检索结果-内蒙古大学图书馆

Filter competition results in more robust Convolutional Neural Networks

NEUROCOMPUTING 2025年 617卷

作者： Gao, Bo Spratling, Michael W. Nanyang Normal Univ Dept Intelligent Mfg & Elect Engn Nanyang 473061 Peoples R China Kings Coll London Dept Informat Strand London WC2R 2LS England Univ Luxembourg Dept Behav & Cognit Sci L-4366 Esch Belval Luxembourg

Convolutional layers, one of the basic building blocks of deep learning architectures, contain numerous trainable filters for feature extraction. These filters operate independently which can result indistinct filters learning similar weights and extracting similar features. In contrast, competition mechanisms in the brain contribute to the sharpening of the responses of activated neurons, enhancing the contrast and selectivity of individual neurons towards specific stimuli, and simultaneously increasing the diversity of responses across the population of neurons. Inspired by this observation, this paper proposes a novel convolutional layer based on the theory of predictive coding, in which each filter effectively tries to block other filters from responding to the input features which it represents. In this way, filters learn to become more distinct which increases the diversity of the extracted features. When replacing standard convolutional layers with the proposed layers the performance of classification networks is not only improved on ImageNet but also significantly boosted on eight robustness benchmarks, as well as on downstream detection and segmentation tasks. Most notably, ResNet50/101/152 robust accuracy increases by 15.9%/20.0%/20.9% under FGSM attack, and by 10.5%/14.7%/15.0% under PGD attack.

关键词： predictive coding Convolutional Neural Networks Robustness

来源：评论

学校读者我要写书评

暂无评论

Task-Oriented Video Compressive Streaming for Real-Time Semantic Segmentation

引用

IEEE TRANSACTIONS ON MOBILE COMPUTING 2024年第12期23卷 14396-14413页

作者： Xiao, Xuedou Zuo, Yingying Yan, Mingxuan Wang, Wei He, Jianhua Zhang, Qian Wuhan Univ Technol Sch Nav Wuhan 430062 Peoples R China Huazhong Univ Sci & Technol Sch Elect Informat & Commun Wuhan 430074 Peoples R China Essex Univ Sch Comp Sci & Elect Engn Colchester CO4 3SQ England Hong Kong Univ Sci & Technol Dept Comp Sci & Engn Clear Water Bay Hong Kong Peoples R China

Real-time semantic segmentation (SS) is a major task for various vision-based applications such as self-driving. Due to the limited computing resources and stringent performance requirements, streaming videos from camera-embedded mobile devices to edge servers for SS is a promising approach. While there are increasing efforts on task-oriented video compression, most SS-applicable algorithms apply more uniform compression, as the sensitive regions are less obvious and concentrated. Such processing results in low compression performance and significantly limits the capacity of edge servers supporting real-time SS. In this paper, we propose STAC, a novel task-oriented DNN-driven video compressive streaming algorithm tailed for SS, to strike accuracy-bitrate balance and adapt to time-varying bandwidth. It exploits DNN's gradients as sensitivity metrics for fine-grained spatial adaptive compression and includes a temporal adaptive scheme that integrates spatial adaptation with predictive coding. Furthermore, we design a new bandwidth-aware neural network, serving as a compatible configuration tuner to fit time-varying bandwidth and content. STAC is evaluated in a system with a commodity mobile device and an edge server with real-world network traces. Experiments show that STAC can save up to 63.7-75.2% of bandwidth or improve accuracy by 3.1-9.5% compared to state-of-the-art algorithms, while capable of adapting to time-varying bandwidth.

关键词： Image coding Bandwidth Streaming media Semantic segmentation Accuracy Servers predictive coding Adaptive streaming DNN-driven compression edge computing semantic segmentation

来源：评论

学校读者我要写书评

暂无评论

Cellular Neural Networks Proposed for Image predictive coding 22

Cellular Neural Networks Proposed for Image Predictive Codin...

引用

22nd International Conference on Nonlinear Dynamics of Electronic Systems (NDES)

作者： Tang, Tang Tetzlaff, Ronald Tech Univ Dresden Fac Elect Engn & Informat Technol Mommsenstr 12 D-01069 Dresden Germany

ISBN: (纸本)9783319086729;9783319086712

In this paper the feasibility of implementing Cellular Neural Networks (CNN) for image predictive coding is investigated. Various CNN structures as predictors are proposed. The performances are compared to the existing predictive coding methods. Thanks to their massive parallel nature, CNN have been proven well suitable for image predictive coding application.

关键词： Cellular Neural Networks predictive coding parallel computing image coding

来源：评论

学校读者我要写书评

暂无评论

Better together: integrating multivariate with univariate methods, and MEG with EEG to study language comprehension

引用

LANGUAGE COGNITION AND NEUROSCIENCE 2024年第8期39卷 991-1019页

作者： Wang, Lin Kuperberg, Gina R. Harvard Med Sch Massachusetts Gen Hosp Dept Psychiat Boston MA USA Harvard Med Sch Massachusetts Gen Hosp Athinoula A Martinos Ctr Biomed Imaging Boston MA USA Tufts Univ Dept Psychol Medford MA USA Harvard Med Sch Massachusetts Gen Hosp Dept Psychiat Boston MA 02129 USA Harvard Med Sch Massachusetts Gen Hosp Athinoula A Martinos Ctr Biomed Imaging Boston MA 02129 USA Tufts Univ Dept Psychol Medford MA 02155 USA

We used MEG and EEG to examine the effects of Plausibility (anomalous vs. plausible) and Animacy (animate vs. inanimate) on activity to incoming words during language comprehension. We conducted univariate event-related and multivariate spatial similarity analyses on both datasets. The univariate and multivariate results converged in their time course and sensitivity to Plausibility. However, only the spatial similarity analyses detected effects of Animacy. The MEG and EEG findings largely converged between 300-500 ms, but diverged in their univariate and multivariate responses to anomalies between 600-1000 ms. We interpret the full set of results within a predictive coding framework. In addition to the theoretical significance, we discuss the methodological implications of the convergence and divergence between the univariate and multivariate results, as well as between the MEG and EEG results. We argue that a deeper understanding of language processing can be achieved by integrating different analysis approaches and techniques.

关键词： ERP representational similarity analysis Animacy predictive coding N400 P600 evidence accumulation

来源：评论

学校读者我要写书评

暂无评论

Learning to segment self-generated from externally caused optic flow through sensorimotor mismatch circuits

引用

NEURAL NETWORKS 2025年 181卷 106716页

作者： Brucklacher, Matthias Pezzulo, Giovanni Mannella, Francesco Galati, Gaspare Pennartz, Cyriel M. A. Univ Amsterdam Cognit & Syst Neurosci NL-1098XH Amsterdam Netherlands CNR Inst Cognit Sci & Technol I-00196 Rome Italy Sapienza Univ Dept Psychol Brain Imaging Lab I-00185 Rome Italy

Efficient sensory detection requires the capacity to ignore task-irrelevant information, for example when optic flow patterns created by egomotion need to be disentangled from object perception. To investigate how this is achieved in the visual system, predictive coding with sensorimotor mismatch detection is an attractive starting point. Indeed, experimental evidence for sensorimotor mismatch signals in early visual areas exists, but it is not understood how they are integrated into cortical networks that perform input segmentation and categorization. Our model advances a biologically plausible solution by extending predictive coding models with the ability to distinguish self-generated from externally caused optic flow. We first show that a simple three neuron circuit produces experience-dependent sensorimotor mismatch responses, in agreement with calcium imaging data from mice. This microcircuit is then integrated into a neural network with two generative streams. The motor-to-visual stream consists of parallel microcircuits between motor and visual areas and learns to spatially predict optic flow resulting from self-motion. The second stream bidirectionally connects a motion-selective higher visual area (mHVA) to V1, assigning a crucial role to the abundant feedback connections to V1: the maintenance of a generative model of externally caused optic flow. In the model, area mHVA learns to segment moving objects from the background, and facilitates object categorization. Based on shared neurocomputational principles across species, the model also maps onto primate visual cortex. Our work extends Hebbian predictive coding to sensorimotor settings, in which the agent actively moves- and learns to predict the consequences of its own movements.

关键词： Generative model Object segmentation predictive coding Optic flow Sensorimotor

来源：评论

学校读者我要写书评

暂无评论

Neurons as autonomous agents: A biologically inspired framework for cognitive architectures in artificial intelligence

引用

COGNITIVE SYSTEMS RESEARCH 2025年 90卷

作者： Luczak, Artur Univ Lethbridge Canadian Ctr Behav Neurosci Lethbridge AB Canada

Despite impressive recent advances in artificial intelligence (AI), current deep neural networks still lack the adaptability and energy efficiency inherent to biological systems. Here we suggest that this problem may be overcome by taking inspiration from the brain where neurons operate as autonomous agents, each capable of adjusting its synaptic connections and internal states based on local information. Currently, typical artificial neurons are static nodes, which is in striking contrast to the rich, dynamic computations performed by biological neurons. In this review, we propose redesigning artificial neurons as self-regulating, agent-like units, making actions to maximize future energy/reward. Similarly, as single-celled organisms which can autonomously navigate in complex environments in search for food, neurons can also be viewed as autonomous decisionmakers, seeking to maximize their own energy resources. Thus, neurons could be operating similarly like reinforcement learning (RL) agents, which make actions to obtain maximum future reward. Here first we review literature illustrating that biological neurons perform complex computations and employ local, predictive learning rules to anticipate future activity to maximize metabolic energy. Next, we provide examples of recent biologically inspired learning algorithms where artificial neurons are empowered with computational flexibility, similarly to autonomous agents. Networks with neurons using such local learning rules can in some examples outperform current AI algorithms. We also discuss how this can improve scalability of current multi-agent systems (MAS) and energy efficiency. Therefore, designing neurons as autonomous agents may provide an important step toward building human-like cognition.

关键词： Biologically inspired AI predictive coding Local learning rules Neuromorphic computing Single-cell intelligence Neuronal plasticity

来源：评论

学校读者我要写书评

暂无评论

Somatosensory false feedback biases emotional ratings through interoceptive embodiment

引用

SCIENTIFIC REPORTS 2025年第1期15卷 1-17页

作者： Patchitt, Joel Garfinkel, Sarah Strawson, William H. Miller, Mark Tsakiris, Manos Clark, Andy Critchley, Hugo D. Univ Sussex Sussex Ctr Consciousness Sci Brighton BN1 9QJ England Univ Sussex Brighton & Sussex Med Sch Dept Clin Neurosci Brighton BN1 9RY England UCL Inst Cognit Neurosci London WC1N 3AZ England Monash Univ Monash Ctr Consciousness & Contemplat Studies Melbourne Vic 3168 Australia Univ Toronto Psychol Dept Toronto ON M5S 2E5 Canada Royal Holloway Univ London Dept Psychol Egham TW20 0EX England Univ Sussex Sch Media Arts & Humanities Brighton BN1 9RG England Univ Sussex Sch Engn & Informat Brighton BN1 9QG England

Mismatches between perceived and veridical physiological signals during false feedback (FFB) can bias emotional judgements. Paradigms using auditory FFB suggest perceived changes in heart rate (HR) increase ratings of emotional intensity irrespective of feedback type (increased or decreased HR), implicating right anterior insula as a mismatch comparator between exteroceptive and interoceptive information. However, few paradigms have examined effects of somatosensory FFB. Participants rated the emotional intensity of randomized facial expressions while they received 20 s blocks of pulsatile somatosensory stimulation at rates higher than HR, lower than HR, equivalent to HR, or no stimulation during a functional magnetic resonance neuroimaging scan. FFB exerted a bidirectional effect on reported intensity ratings of the emotional faces, increasing over the course of each 20 s stimulation block. Neuroimaging showed FFB engaging regions indicative of affective touch processing, embodiment, and reflex suppression. Contrasting higher vs. lower HR FFB revealed engagement of right insula and centres supporting socio-emotional processing. Results indicate that exposure to pulsatile somatosensory stimulation can influence emotional judgements though its progressive embodiment as a perceived interoceptive arousal state, biasing how affective salience is ascribed to external stimuli. Results are consistent with multimodal integration of priors and prediction-error signalling in shaping perceptual judgments.

关键词： Interoception False physiological feedback Insula Emotion recognition Perception predictive coding

来源：评论

学校读者我要写书评

暂无评论

Coiflets, Artificial Neural Networks and predictive coding Based Hybrid Image Compression Methodology 2

Coiflets, Artificial Neural Networks and Predictive Coding B...

引用

2nd International Conference on Devices, Circuits and Systems (ICDCS)

作者： Sridhar, Siripurapu Kumar, P. Rajesh Ramanaiah, K. V. Nataraj, D. Lendi Inst Engn & Technol Jonnada Andhra Pradesh India Andhra Univ Coll Engn Visakhapatnam Andhra Pradesh India Yogi Vemana Univ YSR Engn Coll Kadapa India Pragati Engn Coll Surampalem India

ISBN: (纸本)9781479913565

Hybrid image compression system is discussed and analyzed for better objective fidelity metrics combining the advantages of Coiflet filter functions of wavelets, predictive coding (Differential Pulse Code Modulation-DPCM) and neural networks in addition to quantization and Huffman encoding techniques to eliminate the interpixel, psychovisual redundancy and coding redundancy. Artificial neural networks are self adaptive i.e. they can adjust themselves to data without any specification of the functional model they are fault tolerant by architecture. Wavelets (by choice) on the other hand are computationally simple and provide good compression ratios for high resolution images especially while DPCM removes redundancy in the information. Initially selected wavelet of choice (Coiflet5 in this case) is applied on the input image for two level decomposition generating seven bands of low frequency and high frequency coefficients. The low frequency band 1 coefficients are compressed with DPCM technique while the remaining bands of coefficients are compressed with artificial neural networks. Metrics obtained: Peak Signal to Noise Ratio (PSNR) Mean Square Error (MSE) and Compression Ratio (CR) are tabulated for comparative analysis.

关键词： Coiflets Compression ratio Encoding time predictive coding

来源：评论

学校读者我要写书评

暂无评论

Large-scale interactions in predictive processing: oscillatory versus transient dynamics

引用

TRENDS IN COGNITIVE SCIENCES 2025年第2期29卷 133-148页

作者： Vinck, Martin Uran, Cem Dowdall, Jarrod R. Rummell, Brian Canales-Johnson, Andres Max Planck Gesell Ernst Strungmann Inst ESI Neurosci D-60528 Frankfurt Germany Radboud Univ Nijmegen Donders Ctr Neurosci Dept Neurophys NL-6525 Nijmegen Netherlands Western Univ Robarts Res Inst London ON Canada Univ Catolica Maule Fac Ciencias Salud Talca 3480122 Chile Univ Cambridge Dept Psychol Cambridge CB2 3EB England

How do the two main types of neural dynamics, aperiodic transients and oscillations, contribute to the interactions between feedforward (FF) and feedback (FB) pathways in sensory inference and predictive processing? We discuss three theoretical perspectives. First, we critically evaluate the theory that gamma and alpha/beta rhythms play a role in classic hierarchical predictive coding (HPC) by mediating FF and FB communication, respectively. Second, we outline an alternative functional model in which rapid sensory inference is mediated by aperiodic transients, whereas oscillations contribute to the stabilization of neural representations over time and plasticity processes. Third, we propose that the strong dependence of oscillations on predictability can be explained based on a biologically plausible alternative to classic HPC, namely dendritic HPC.

关键词： predictive processing predictive coding classical hierarchical predictive coding dendritic hierarchical predictive coding rhythms oscillations synchronization dynamics gamma alpha beta feedforward feedback hierarchy efficient coding plasticity

来源：评论

学校读者我要写书评

暂无评论

The selective updating of working memory: a predictive coding account

The selective updating of working memory: a predictive codin...

引用

作者： Yen Yu University of London

学位级别：博士

Goal-relevant information maintained in working memory is remarkably robust and resistant to distractions. However, our nervous system is endowed with exceptional flexibility; therefore such information can be updated almost effortlessly. A scenario – not uncommon in our daily life – is that selective maintaining and updating information can be achieved concurrently. This is an intriguing example of how our brain balances stability and flexibility, when organising its knowledge. A possibility – one may draw upon to understand this capacity – is that working memory is represented as beliefs, or its probability densities, which are updated in a context-sensitive manner. This means one could treat working memory in the same way as perception – i.e., memories are based on inferring the cause of sensations, except that the time scale ranges from an instant to prolonged anticipation. In this setting, working memory is susceptible to prior information encoded in the brain’s model of its world. This thesis aimed to establish an interpretation of working memory processing that rests on the (generalised) predictive coding framework, or hierarchical inference in the brain. Specifically, the main question it asked was how anticipation modulates working memory updating (or maintenance). A novel working memory updating task was designed in this regard. Blood-oxygen-level dependent (BOLD) imaging, machine learning, and dynamic causal modelling (DCM) were applied to identify the neural correlates of anticipation and the violation of anticipation, as well as the causal structure generating these neural correlates. Anticipation induced neural activity in the dopaminergic midbrain and the striatum. Whereas, the fronto-parietal and cingulo-operculum network were implicated when an anticipated update was omitted, and the midbrain, occipital cortices, and cerebellum when an update was unexpected. DCM revealed that anticipation is a modulation of backward connections, whilst the associate

关键词： working memory updating predictive coding fMRI dynamic causal modelling MVPA

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：