Method, Digital Electronic Circuit and System for Unsupervised Detection of Repeating Patterns in a Series of Events

Abstract

A method of performing unsupervised detection of repeating patterns in a series (TS) of events (E21, E12, E5 . . . ), comprising the steps of: a) Providing a plurality of neurons (NRI-NRP), each neuron being representative of W event types; b) Acquiring an input packet (IV) comprising N successive events of the series; c) Attributing to at least some neurons a potential value (PTI-PTP), representative of the number of common events between the input packet and the neuron; d) Modify the event types of neurons having a potential value exceeding a first threshold TL; and e) generating a first output signal (OSI-OSP) for all neurons having a potential value exceeding a second threshold TF, and a second output signal, different from the first one, for all other neurons. A digital electronic circuit and system configured for carrying out such a method.

What is claimed is: 
     
         1 . A digital electronic circuit architecture for unsupervised detection of repeating patterns, the digital electronic circuit comprising:
 an input port configured to receive a first series of digital packets representing respective events;   a memory configured to store data defining a first layer having a plurality of neurons, where each of the plurality of neurons is associated with a set of binary weights; and   a processing circuit configured to:
 output a second series of digital packets indicative of neurons in the plurality of neurons having a potential value exceeding a firing threshold value; and 
 modify binary weights associated with neurons having a potential value exceeding a learning threshold value, wherein the modifying swaps binary weights that are mismatched with a current digital packet with other binary weights based on signal types within the first series of digital packets. 
   
     
     
         2 . The digital electronic circuit architecture of  claim 1 , wherein the firing threshold value is a same value across the first layer. 
     
     
         3 . The digital electronic circuit architecture of  claim 1 , wherein the firing threshold value is set based on a hyper-geometric distribution. 
     
     
         4 . The digital electronic circuit architecture of  claim 1 , wherein a number of binary weights in the set of binary weights is a same number for each of the plurality of neurons. 
     
     
         5 . The digital electronic circuit architecture of  claim 1 , wherein the learning threshold value is updated during learning by the processing circuit. 
     
     
         6 . The digital electronic circuit architecture of  claim 1 , wherein a number of available weights is updated during learning by the processing circuit. 
     
     
         7 . The digital electronic circuit architecture of  claim 1 , wherein the processing circuit is further configured to compute the potential value of each neuron by incrementing a counter after applying an AND function to the binary values in the first series of digital packets and the binary weights to generate results. 
     
     
         8 . The digital electronic circuit architecture of  claim 1 , wherein the memory is further configured to store data defining a second layer having a second plurality of neurons, where each of the second plurality of neurons is associated with a second set of binary weights, the second plurality of neuron receiving input from the outputted second series of digital packets from the first layer. 
     
     
         9 . The digital electronic circuit architecture of  claim 8 , wherein a packet length associated with the first layer differs from a packet length associated with the second layer. 
     
     
         10 . The digital electronic circuit architecture of  claim 1 , wherein the first series of digital packets are derived from a video stream or a set of images. 
     
     
         11 . A method for performing unsupervised detection of repeating patterns, the method comprising:
 receiving, via an input port of a digital electronic circuit architecture, a first series of digital packets representing respective events, the digital electronic circuit architecture including a memory configured to store data defining a first layer having a plurality of neurons, where each of the plurality of neurons is associated with a set of binary weights;   outputting, by a processing circuit, a second series of digital packets indicative of neurons in the plurality of neurons having a potential value exceeding a firing threshold value; and   modifying, by the processing circuit, binary weights associated with neurons having a potential value exceeding a learning threshold value, wherein the modifying swaps binary weights that are mismatched with a current digital packet with other binary weights based on signal types within the first series of digital packets.   
     
     
         12 . The method of  claim 11 , wherein the firing threshold value is a same value across the first layer. 
     
     
         13 . The method of  claim 11 , wherein the firing threshold value is set based on a hyper-geometric distribution. 
     
     
         14 . The method of  claim 11 , wherein a number of binary weights in the set of binary weights is a same number for each of the plurality of neurons. 
     
     
         15 . The method of  claim 11 , wherein the learning threshold value is updated during learning by the processing circuit. 
     
     
         16 . The method of  claim 11 , wherein a number of available weights is updated during learning by the processing circuit. 
     
     
         17 . The method of  claim 11 , further comprising:
 computing, by the processing circuit, the potential value of each neuron by incrementing a counter after applying an AND function to the binary values in the first series of digital packets and the binary weights to generate results.   
     
     
         18 . The method of  claim 11 , wherein the memory is further configured to store data defining a second layer having a second plurality of neurons, where each of the second plurality of neurons is associated with a second set of binary weights, the second plurality of neuron receiving input from the outputted second series of digital packets from the first layer. 
     
     
         19 . The method of  claim 18 , wherein a packet length associated with the first layer differs from a packet length associated with the second layer. 
     
     
         20 . The method of  claim 11 , wherein the first series of digital packets are derived from a video stream or a set of images.