网范文：“ Sparse visual models for sensorimotor control” 在生存竞争中鉴于有效使用资源的重要性,有理由认为自然进化已经发现自然的视觉信息。具有内在优势的容错观念和低功耗消费潜力,关于感觉运动控制与强大的决策部署。灵感来自于哺乳动物大脑和视觉通路,在这篇生物范文中，英语论文范文，提出层次稀疏编码，提取视觉特性的网络体系结构用于感觉运动控制。测试与自然图像表明,其编码便于处理和学习。
先前的探讨已经表明复杂细胞的反应可能表示为一个高阶神经层。这里我们扩展编码在每个网络层,显示早期视觉的详细建模，英语论文题目，提高开发的后续阶段。下面的范文进行详述。
Abstract
Given the importance of using resources efficiently in the competition for survival, it is reasonable to think that natural evolution has discovered efficient cortical coding strategies for representing natural visual information. Sparse representations have intrinsic advantages in terms of fault-tolerance and low-power consumption potential, and can therefore be attractive for robot sensorimotor control with powerful dispositions for decisionmaking. Inspired by the mammalian brain and its visual ventral pathway, we present in this a hierarchical sparse coding network architecture that extracts visual features for use in sensorimotor control. Testing with natural images demonstrates that this sparse coding facilitates processing and learning in subsequent layers. Previous studies have shown how the responses of complex cells could be sparsely represented by a higher-order neural layer. Here we extend sparse coding in each network layer, showing that detailed modeling of earlier stages in the visual pathway enhances the characteristics of the receptive fields developed in subsequent stages. The yield network is more dynamic with richer and more biologically plausible input and output representation.
Introduction
One of the major difficulties in robot navigation is the capability of rapidly responding to unpredictable and novel situations. The level of delegation and autonomy of robotic systems in remote missions will depend on their computational capabilities for decision-making, their adaptation and learning, and their capability to survive uncertain environments. The engineering of such systems will be more complex than ever, and design faults can be very costly.
Resilience to design faults, faults due to the elements, and proper responses to novel situations will be critical. Biological systems provide solutions to similar problems, and the brain provides a rich source of computational paradigms that can be used as inspiration for revolutionary computational solutions (Mousset, Jabri et al. 2017; Jabri 2017; Wang, Jin et al. 2017). We are developing in our laboratory biologically inspired sensorimotor controlsystems for controlling robots and decisions systems. Our approach to visual processing is based on the modeling of the visual ventral pathway, and this is the focus of the present .
The mammalian cortex has evolved over millions of years to effectively cope with visual information of the natural environment. Given the importance of using resources efficiently in the competition for survival, it is reasonable to think that natural evolution has discovered efficient cortical coding strategies for representing natural visual information. Here, the notion of efficiency is based on Barlow’s principle of redundancy reduction (Barlow 1994), which proposes that a useful goal of sensor coding is to transform the input in a manner that reduces the redundancy due to complex statistical dependencies among elements of the input streams. The usefulness of redundancy reduction can be understood by considering the process of image formation, which occurs by light reflecting off of independent entities (i.e. objects) in the world and being focused onto an array of photoreceptors in the retina.
The activities of the photoreceptors themselves do not form a particularly useful signal to the organism because the structure present in the world is not made explicit, but rather is embedded in the form of complex statistical dependencies, or redundancies among photoreceptor activities. A reasonable goal of the visual system is to extract these statistical dependencies so that images may be explained in terms of a collection of independent events, so that means forming the sparse representation for a given image. The hope is that such a sparse coding strategy will recover an explicit representation of the underlying independent entities that gave rise to the image, which would be useful to the survival of organism.
Furthermore, sparse coding (Amari 1993) has been proven to provide superior information storage capacity compared to local (grandmother cell theory or Gnostic representation) or distributed information representations. Because only very few neurons need to be activated and that there are only a few neurons encoding an event, sparse representation have intrinsic fault-tolerance and low-power consumption potential. Fault tolerance is a critical requirement in the remote deployment of intelligent systems, which has been attributed to neural networks because of the distributed representations that develop during learning.
Another important aspect of any physical realization of computational models is the power consumed. Biologically based principles such as sparse representation may have the information processing capabilities as well as huge payoffs in power/energy minimization and optimal resource management. Also, sparse representations are important from an implementation perspective. The physical connectivity of large scale networks require strategies that exploits sparseness of networks, local connectivity, population-based encoding, and information flow in operation as well as during learning.
Visual processing framework For sensorimotor control, individual landmarks and goal locations must be extracted from complex visual scenes, and objects and their spatial relationships must be identified. Generally, the real world almost always contains more information than we can process at any given point in time, so we must learn to use it iteratively, searching for the most relevant information at any given point in time. On the other hand, findings from neurophysiology, psychophysics, and fMRI (Reynolds, Chelazzi et al. 1999; Kanwisher and Wojciulik 2017; Reynolds, Pasternak et al. 2017) all point to the roles of attention and stimulus salience in biasing the competition of neurons in ventral stream to facilitate object recognition. When subjects are instructed to attend (or choose voluntarily to attend) to a stimulus at a particular location or with a particular feature, this generates signals within areas outside visual cortex, such as parietal cortex, frontal eye field (FEF), prefrontal cortex, and amygadala. These signals are then fed back to extrastriate areas, where they bias the competition in these areas in favor of neurons that respond to the features or location of the attended stimulus. As a result, neurons that respond to the attended stimulus remain active while suppressing neurons that respond to the ignored stimuli. In other words, neuronal responses are now determined by the attended stimulus. In absence of attention control, the most salient element in the scene might dominate neuronal responses.
In this visual processing system, the bottom-up salience and top-down attention can complementarily filter out unwanted information from typically cluttered real-world scenes and to focus on what is important in a given situation. This will largely reduce the computational complexity and simplified the object recognition process.
Discussion
We presented in this a hierarchical network architecture inspired by the mammalian ventral pathway to sparsely represent visual features for use in sensorimotor control. This sparse representation provided intrinsic low power and fault-tolerant computing substrate to sensorimotor control systems. By unsupervised learning algorithms, the learned visual models made the sensorimotor control systems to automatically adapt to uncertain and novel environment. We also show that in such a model, V2 cell receptive fields develop end-stopping properties. According to Hubel and Wiesel, the optimal stimulus for an end-stopped cell is a line that extends for a certain distance and no further. For a cell that responds to edges and is endstopped at one end only, a corner is ideal; for a cell that responds to slits or black bars and is stopped at both ends, the optimum stimulus is a short white or black line or a line that curves so that it is appropriate in the activating region and inappropriate (different by 20 to 30 degrees or more) in flanking regions. We can thus view end-stopped cells as sensitive to corners, to curvature, or to sudden breaks in line. These contours are very crucial for shape representation in cortex V4 (Gallant, Braun et al. 1993; Wilkinson, James et al. 2017; Pasupathy and Connor 2017), thus they are very important for object representation and recognition in IT.()
网站原创范文除特殊说明外一切图文作品权归所有；未经官方授权谢绝任何用途转载或刊发于媒体。如发生侵犯作品权现象，保留一切法学追诉权。()
更多范文欢迎访问我们主页 当然有需求可以和我们 联系交流。-X()
|
