Granted Patent B2

US 12,367,393 · App. 18/656,669 · Granted Jul 22, 2025

Inventors: Henry Markram (Lausanne, CH); Felix Schuermann (Grens, CH); John Rahmon (Lausanne, CH); Daniel Milan Lütgehetmann (Lausanne, CH); Constantin Cosmin Atanasoaei (Chavannes-pres-Renens, CH)

Assignee: INAIT SA

G06N3/08G06N3/044G06N3/049

Abstract

Claims (29)

1. A method comprising:

defining a first window of time and a second window of time in a spiking recurrent artificial neural network, wherein the first window of time has a shorter duration than a duration of the second window of time;

inputting data into the spiking recurrent artificial neural network, wherein the input data comprises first data that originates from a first sensor and second data that originates from a second sensor, wherein the first data occurs over a shorter duration than the second data;

identifying occurrences of topological patterns of activity in a functional graph of the spiking recurrent artificial neural network during the first window of time, wherein the activity during the first window of time is responsive to the input data;

identifying occurrences of topological patterns of activity in the functional graph of the spiking recurrent artificial neural network during the second window of time, wherein the activity during the second window of time is responsive to the input data; and

outputting representations of the identified occurrences of topological patterns of activity identified during both the first window of time and the second window of time.

2. The method of claim 1 , further comprising identifying, using the output representations, that a first of the topological patterns of activity that occurred in the first window of time reoccurred during the first window.

3. The method of claim 2 , further comprising adjusting one or more characteristics of the spiking recurrent artificial neural network to reinforce occurrence of the first of the topological patterns in a subsequent representation output from the spiking recurrent artificial neural network, wherein the subsequent representation represents identified occurrences of topological patterns of activity responsive to the input data and identified during the first window of time and the second window of time.

4. The method of claim 1 , further comprising identifying that a first of the topological patterns of activity that occurred during the first window does not reoccur in the second window.

5. The method of claim 4 , further comprising eliminating a representation of the first of the topological patterns in a subsequent representation output from the spiking recurrent artificial neural network.

6. The method of claim 4 , further comprising adjusting one or more characteristics of the spiking recurrent artificial neural network to attenuate the occurrence of the first topological pattern during the first window in a subsequent representation output from the spiking recurrent artificial neural network, wherein the subsequent representation represents identified occurrences of topological patterns of activity responsive to the input data and identified during the first window of time and the second window of time.

7. The method of claim 6 , wherein adjusting the one or more characteristics of the spiking recurrent artificial neural network comprises adding or removing either nodes or links from the spiking recurrent neural network.

8. The method of claim 6 , wherein adjusting the one or more characteristics of the spiking recurrent artificial neural network comprises changing a weights of links in the spiking recurrent artificial neural network.

9. The method of claim 1 , wherein:

the first data includes a still image or a temperature measurement; and

the second data includes video data or audio data.

10. The method of claim 1 , wherein the topological patterns of activity comprise directed cliques or directed simplices.

11. The method of claim 1 , further comprising comparing a first of the representations of the occurrences of the topological patterns of activity that occurred during the first window with a second of the representations of the occurrences of the topological patterns of activity that occurred in the second window of time.

12. The method of claim 11 , wherein comparing the first and the second representations comprises comparing values that reflect a level of activity of the represented topological patterns.

13. The method of claim 11 , wherein comparing the first and the second representations comprises comparing values that reflect a duration of the represented topological patterns.

14. The method of claim 11 , further comprising classifying, based on the comparison,

a first decision that is represented by the first representation as less robust than

a second decision that is represented by the second representation.

15. The method of claim 11 , further comprising:

classifying, based on a result of the comparison, a first decision that is represented by the first representation as a result of an adversarial attack.

16. The method of claim 11 , wherein comparing the first and the second representations comprises:

subtracting a first collection of binary digits from a second collection of binary digits, wherein each binary digit indicates whether a respective topological pattern occurred.

17. The method of claim 1 , wherein the first window starts before the second window.

18. The method of claim 1 , wherein the first data and the second data are either first and second images that feature a same subject or first and second text snippets that share a textual characteristic.

Assignments (1)

Continuity (3)

References Cited (400)