公开(公告)号：WO2015142503A3

公开(公告)日：2015-12-23

申请号：PCT/US2015018264

申请日：2015-03-02

Applicant: QUALCOMM INC

Inventor： LEVIN JEFFREY ALEXANDER ,  MALONE ERIK CHRISTOPHER ,  LIAO EDWARD HANYU

IPC: G06N3/04 ,  G06N3/063

CPC classification number: G06N3/08 ,  G06N3/049 ,  G06N3/063

Abstract: Certain aspects of the present disclosure support a method and apparatus for implementing kortex neural network processor within an artificial nervous system. According to certain aspects, a plurality of spike events can be generated by a plurality of neuron unit processors of the artificial nervous system, and the spike events can be sent from a subset of the neuron unit processors to another subset of the neuron unit processors via a plurality of synaptic connection processors of the artificial nervous system.

Abstract translation: 本公开的某些方面支持用于在人造神经系统内实现皮质神经网络处理器的方法和设备。 根据某些方面，多个尖峰事件可以由人造神经系统的多个神经元单元处理器产生，并且尖峰事件可以经由神经元单元处理器的子集经由神经元单元处理器的子集经由 人造神经系统的多个突触连接处理器。

公开(公告)号：WO2015088774A2

公开(公告)日：2015-06-18

申请号：PCT/US2014067343

申请日：2014-11-25

Applicant: QUALCOMM INC

Inventor： GUPTA VIKRAM ,  LEVIN JEFFREY ALEXANDER ,  LIAO EDWARD HANYU

IPC: G06N3/04

CPC classification number: G06N3/08 ,  G06N3/049

Abstract: A method for pattern recognition in a spiking neural network robust to initial network conditions includes creating a set of diverse neurons in a first layer to increase a diversity in a set of spike timings. An input corresponding to a pattern plus noise is presented at an input layer and represented as spikes. The spikes are received at the first layer and spikes are produced at the first layer based on the received spikes. The method also includes updating a weight of each synapse between an input layer neuron and an output layer neuron based on a spike timing difference between a spike at the input layer neuron and a spike at the output layer neuron. Further, the method includes classifying a spike pattern represented by a set of inter-spike intervals, regardless of noise in the spike pattern.

Abstract translation: 在对初始网络条件稳健的加标神经网络中的模式识别的方法包括在第一层中创建一组不同的神经元以增加一组尖峰定时的分集。 对应于图案加噪声的输入在输入层处呈现并表示为尖峰。 尖峰在第一层被接收，并且基于接收的尖峰在第一层产生尖峰。 该方法还包括基于输入层神经元的尖峰与输出层神经元的尖峰之间的尖峰定时差来更新输入层神经元和输出层神经元之间的每个突触的权重。 此外，该方法包括分类由一组间穗间隔表示的尖峰图案，而不管尖峰图案中的噪声如何。