Neural Implementation of Shape-Invariant Touch Counter Based on Euler Calculus
Neural Implementation of Shape-Invariant Touch Counter Based on Euler Calculus
复制标题
DOI：
10.1109/access.2014.2351832
复制
发表时间：
2014-08
期刊：
影响因子：
3.9
通讯作者：
K. Miura;K. Nakada
中科院分区：
文献类型：
--
作者：
K. Miura;K. Nakada
神经形态工程的目标之一是模仿大脑的能力，识别和计算作为实体的个体物体的数量，这是基于来自激活的触觉（或视觉）感觉神经元群体的信息的全局一致性，无论物体的形状如何。为了实现这种灵活性，可能值得研究一种非传统的算法，如拓扑方法。在这里，我们提出了一个完全并行化的算法的形状不变的触摸计数器的2-D像素。触摸的数量通过欧拉积分（一种广义积分）来计数，其中用于二进制图像的连通分量计数器（贝蒂数）被用作基本模块。通过触摸的例子，我们清楚地展示了所提出的电路架构如何以递归神经网络的形式体现欧拉积分，以进行迭代向量运算。我们的并行化可以导致现场可编程门阵列或数字信号处理器实现的拓扑算法的可扩展性，以高分辨率的像素。
One of the goals of neuromorphic engineering is to imitate the brain's ability to recognize and count the number of individual objects as entities based on the global consistency of the information from the population of activated tactile (or visual) sensory neurons whatever the objects' shapes are. To achieve this flexibility, it may be worth examining an unconventional algorithm such as topological methods. Here, we propose a fully parallelized algorithm for a shape-invariant touch counter for 2-D pixels. The number of touches is counted by the Euler integral, a generalized integral, in which a connected component counter (Betti number) for the binary image was used as elemental module. Through examples of touches, we demonstrate transparently how the proposed circuit architecture embodies the Euler integral in the form of recurrent neural networks for iterative vector operations. Our parallelization can lead the way to Field-Programmable Gate Array or Digital Signal Processor implementations of topological algorithms with scalability to high resolutions of pixels.