# US 2019/0377976 A1 · Characterizing activity in a recurrent artificial neural network

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0377976 A1
- Document ID: US20190377976A1
- Status: Published application
- Kind: A1
- Application number: 16/004,837
- Filed: 2018-06-11
- Published: 2019-12-12
- Earliest claimed: 2018-06-11
- Assignees: Inait SA
- Applicants: Inait SA
- Inventors: Henry Markram; Ran Levi; Kathryn Pamela Hess Bellwald
- Web page: https://patentgrep.com/patent/US20190377976A1

## CPC

- G06F 18/2413: based on distances to training or reference patterns
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06F 18/2185: {the supervisor being an automated module, e.g. intelligent oracle}
- G06F 18/285: {Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/045: Combinations of networks
- G06N 3/09: Supervised learning
- G06N 3/092: Reinforcement learning
- G06N 7/08: using chaos models or non-linear system models
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/82: using neural networks
- G06F 11/3003: {Monitoring arrangements specially adapted to the computing system or computing system component being monitored}
- G06F 2218/12: Classification; Matching
- G06V 20/00: Scenes; Scene-specific elements (control of digital cameras H04N23/60)
- G06V 20/44: {Event detection}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for characterizing activity in a recurrent artificial neural network. In one aspect, a method for identifying decision moments in a recurrent artificial neural network includes determining a complexity of patterns of activity in the recurrent artificial neural network, wherein the activity is responsive to input into the recurrent artificial neural network, determining a timing of activity having a complexity that is distinguishable from other activity that is responsive to the input, and identifying the decision moment based on the timing of the activity that has the distinguishable complexity.

## Claims

### Claim 1

A method for identifying decision moments in a recurrent artificial neural network, the method comprising:
determining a complexity of patterns of activity in the recurrent artificial neural network;
determining a timing of activity having a complexity that is distinguishable from complexity of other activity that is responsive to the input; and
identifying a decision moment based on the timing of the activity that has the distinguishable complexity.

### Claim 2 (depends on 1)

The method of claim 1, wherein determining the complexity of patterns comprises determining a number of patterns enclosing cavities.

### Claim 3 (depends on 1)

The method of claim 1, further comprising estimating whether the activity is responsive to the input into the recurrent artificial neural network, the estimating comprising:
estimating that relatively simpler patterns of activity relatively soon after the input event are responsive the input but that relatively more complex patterns of activity relatively soon after the input event are not responsive the input; and
estimating that relatively more complex patterns of activity relatively later after the input event are responsive the input but that relatively simpler patterns of activity relatively later after the input event are not responsive the input.

### Claim 4 (depends on 1)

The method of claim 1, wherein determining the timing of the activity having the distinguishable complexity comprises determining the timing of activity having relatively high complexity indicative of ordered activity within the recurrent artificial neural network, wherein the relatively high complexity is determined relative to a baseline level of activity or to an average level of activity responsive to a plurality of different inputs.

### Claim 5 (depends on 1)

The method of claim 1, wherein determining the complexity of the patterns of activity comprises identifying clique patterns of activity of the recurrent artificial neural network.

### Claim 6 (depends on 3)

The method of claim 3, wherein the method further comprises defining a plurality of windows of time during which the activity of the recurrent artificial neural network is responsive to an input into the recurrent artificial neural network, wherein the clique patterns of activity are identified in each of the pluralities of windows of time.

### Claim 7 (depends on 4)

The method of claim 4, wherein the method further comprises identifying a first window of time within the plurality of windows of time based on a distinguishable likelihood of the clique patterns of activity being identified during the first window.

### Claim 8 (depends on 3)

The method of claim 3, wherein identifying clique patterns comprises identifying groups of directed cliques that enclose cavities.

### Claim 9 (depends on 5)

The method of claim 5, wherein determining the complexity of the patterns of activity comprises weighting a number of occurrences of the clique patterns according to a number of points in the respective clique patterns.

### Claim 10 (depends on 5)

The method of claim 5, wherein determining the complexity of the patterns of activity comprises weighting a number of occurrences of the clique patterns according to a directionality of the respective clique patterns.

### Claim 11 (depends on 1)

The method of claim 1, wherein:
determining the complexity of the patterns of activity comprises
dividing activity in the recurrent artificial neural network and responsive to input into the recurrent artificial neural network into a plurality of windows of time, and
determining the complexity of the patterns of the activity within each of the windows of time; and
determining the timing of the activity having the distinguishable complexity comprises identifying a first window of time having the distinguishable complexity.

### Claim 12 (depends on 1)

The method of claim 1, wherein identifying the decision moment comprises identifying times when the activity has a relatively high complexity indicative of ordered activity within the recurrent artificial neural network.

### Claim 13 (depends on 1)

The method of claim 1, further comprising reading one or more outputs of the recurrent artificial neural network at the decision moment, wherein the outputs are based on the patterns of activity indicative of ordered activity within the recurrent artificial neural network.

### Claim 14 (depends on 13)

The method of claim 13, wherein the outputs a collection of multi-valued, non-binary digits, wherein the values each represent a weight assigned to edges in a corresponding of the patterns of activity.

### Claim 15 (depends on 13)

The method of claim 13, further comprising transmitting or storing at least some of the read outputs, wherein transmitting or storing at least some of the read outputs comprises:
transmitting or storing outputs associated with patterns of activity having a relatively higher complexity; and
discarding or ignoring outputs associated with patterns of activity having a relatively lower complexity.

### Claim 16 (depends on 1)

The method of claim 1, wherein the input is a continuous stream of data input into the recurrent artificial neural network.

### Claim 17

A system comprising one or more computers operable to perform operations, the operations comprising:
determining a complexity of patterns of activity in the recurrent artificial neural network;
determining a timing of patterns of activity having a complexity that is distinguishable from complexity of other patterns of activity; and
identifying a decision moment based on the timing of the patterns that have the distinguishable complexity.

### Claim 18 (depends on 17)

The system of claim 17, wherein determining the complexity of patterns comprises determining a number of patterns enclosing cavities.

### Claim 19 (depends on 17)

The system of claim 17, wherein the activity is responsive to input into the recurrent artificial neural network.

### Claim 20 (depends on 17)

The system of claim 17, wherein determining the timing of the patterns having the distinguishable complexity comprises determining the timing of patterns having relatively high complexity indicative of ordered activity within the recurrent artificial neural network, wherein the relatively high complexity is determined relative to a baseline level of pattern complexity or to an average level of pattern complexity responsive to a plurality of different inputs.

### Claim 21 (depends on 17)

The system of claim 17, wherein determining the complexity of the patterns of activity comprises identifying pre-defined clique patterns of activity of the recurrent artificial neural network.

### Claim 22 (depends on 21)

The system of claim 21, wherein the method further comprises defining a plurality of windows of time during which the activity of the recurrent artificial neural network is responsive to an input into the recurrent artificial neural network, wherein the pre-defined clique patterns of activity are identified in each of the pluralities of windows of time.

### Claim 23 (depends on 22)

The system of claim 22, wherein the method further comprises identifying a first window of time within the plurality of windows of time based on a distinguishable likelihood of the pre-defined clique patterns of activity being identified during the first window.

### Claim 24 (depends on 21)

The system of claim 21, wherein identifying clique patterns comprises identifying groups of directed cliques that enclose cavities.

### Claim 25 (depends on 21)

The system of claim 21, wherein determining the complexity of the patterns of activity comprises weighting a number of occurrences of the clique patterns according to a number of points in the respective clique patterns.

### Claim 26 (depends on 21)

The system of claim 21, wherein determining the complexity of the patterns of activity comprises weighting a number of occurrences of the clique patterns according to a directionality of the respective clique patterns.

### Claim 27 (depends on 17)

The system of claim 17, wherein:
determining the complexity of the patterns of activity comprises
dividing activity in the recurrent artificial neural network and responsive to input into the recurrent artificial neural network into a plurality of windows of time, and
determining the complexity of the patterns of the activity within each of the windows of time; and
determining the timing of the patterns of activity having the distinguishable complexity comprises identifying a first window of time during which the patterns have the distinguishable complexity.

### Claim 28 (depends on 17)

The system of claim 17, wherein identifying the decision moment comprises identifying times when the patterns of activity have encompass cavities.

### Claim 29 (depends on 17)

The system of claim 17, further comprising reading one or more outputs of the recurrent artificial neural network at the decision moment, wherein the outputs are based on the patterns of activity indicative of ordered activity within the recurrent artificial neural network.

### Claim 30 (depends on 29)

The system of claim 29, wherein the outputs comprise a collection of multi-valued, non-binary digits, wherein the values each represent a weight assigned to edges in a corresponding of the patterns of activity.

### Claim 31 (depends on 29)

The system of claim 29, further comprising transmitting or storing at least some of the read outputs, wherein transmitting or storing at least some of the read outputs comprises:
transmitting or storing outputs associated with patterns of activity having a relatively higher complexity; and
discarding or ignoring outputs associated with patterns of activity having a relatively lower complexity.

### Claim 32 (depends on 17)

The system of claim 17, wherein the operations further comprise estimating whether the activity is responsive to the input into the recurrent artificial neural network, the estimating comprising:
estimating that relatively simpler patterns of activity relatively soon after the input event are responsive the input but that relatively more complex patterns of activity relatively soon after the input event are not responsive the input; and
estimating that relatively more complex patterns of activity relatively later after the input event are responsive the input but that relatively simpler patterns of activity relatively later after the input event are not responsive the input.
