# US 2019/0379692 A1 · Method for predicting events using a joint representation of different feature types

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

- Number: US 2019/0379692 A1
- Document ID: US20190379692A1
- Status: Published application
- Kind: A1
- Application number: 16/480,318
- Filed: 2017-12-01
- Published: 2019-12-12
- Earliest claimed: 2017-02-17
- Assignees: NEC Laboratories Europe GmbH
- Applicants: NEC Laboratories Europe GmbH
- Inventors: Alberto Garcia Duran; Mathias Niepert
- Web page: https://patentgrep.com/patent/US20190379692A1

## CPC

- H04L 63/1441: {Countermeasures against malicious traffic (countermeasures against attacks on cryptographic mechanisms H04L9/002)}
- G06F 21/552: {involving long-term monitoring or reporting}
- G06N 3/0499: Feedforward networks
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- H04L 63/1416: {Event detection, e.g. attack signature detection}
- H04L 67/535: {Tracking the activity of the user (network monitoring arrangements H04L43/00; recording of computer activity G06F11/34)}
- G06F 2221/2111: Location-sensitive, e.g. geographical location, GPS
- H04L 2463/146: Tracing the source of attacks
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for predicting one or more events includes generating, for features of each of at least two feature types, an intermediate representation using a representation learning model for the at least two feature types. The intermediate representations of the at least two feature types are analyzed using a neural network and at least one neural network model so as to provide a joint representation for predicting certain events. One or more actions to be taken can be determined based on the one or more events predicted by the joint representation.

## Claims

### Claim 1

A method for predicting one or more events, the method comprising:
a) generating, for features of each of at least two feature types, an intermediate representation using a representation learning model for the at least two feature types,
b) analyzing the intermediate representations of the at least two feature types using a neural network and at least one neural network model so as to provide a joint representation for predicting certain events, and
c) determining one or more actions to be taken based on the one or more events predicted by the joint representation.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the at least two feature types include device features and user features, and wherein the device features and/or the user features are preprocessed to provide the features in a predefined format for further processing.

### Claim 3 (depends on 1)

The method according to claim 1, wherein request features of requests are preprocessed to identify unique or consecutive requests.

### Claim 4 (depends on 1)

The method according to claim 1, wherein the one or more actions are for operating a network and the one or more actions are translated into one or more operations for operating the network.

### Claim 5 (depends on 1)

The method according to claim 1, wherein past events and representations of features associated with the past events are stored in a database connected to the neural network and used to train the neural network.

### Claim 6 (depends on 1)

The method according to claim 1, wherein the at least two feature types include device features and user features, and wherein the device features are analyzed only if one or more of values of the devices features are within a predetermined range.

### Claim 7 (depends on 4)

The method according to claim 4, wherein said the one or more events are security events and said the one or more operations are operations for countermeasures against a breach of security related to the security events.

### Claim 8 (depends on 1)

The method according to claim 1, wherein predefined request attributes are used for generating a request key, the request key being used for preprocessing of requests to identify unique requests.

### Claim 9 (depends on 8)

The method according to claim 8, wherein consecutive ones of the requests having a same request key are merged into a common single request.

### Claim 10 (depends on 1)

The method according to claim 1, wherein:
a request feature of a request specifies a domain type, a content type, a protocol type or the like;
a device feature of a device specifies a geolocation, a power-consuming pattern, on-off-pattern, data volume sent and/or received, data traffic patterns for sent or received data of the device or the like; and/or
a user feature of a user specifies place of birth, an address, age, current position, height color of hair or eyes, most visited places, most used train stations or the like.

### Claim 11

(canceled)

### Claim 12

A neural network entity for predicting certain events, comprising:
an input interface configured to receive intermediate representations for features of each of at least two feature types which were learned using a representation learning model for the at least two feature types;
a processing component connected to the input interface and configured to analyze the intermediate representations for the features by using at least one neural network model to generate a joint representation for predicting the certain events, based on a result of the analysis; and
an output interface configured to provide the joint representation for use in predicting an existence or occurrence of at least one of the certain events.

### Claim 13

A method for operating a neural network entity for predicting certain events, the method comprising:
a) receiving, by an input interface, intermediate representations for features of each of at least two feature types which were learned using a representation learning model for the at least two feature types;
b) analyzing, by a processing component of the neural network entity connected to the input interface, the intermediate representations for the features using at least one neural network model;
c) generating, by the processing component, a joint representation for predicting the certain events, based on the result of the analysis of step b); and
d) providing, by an output interface, the joint representation for use in predicting an existence or occurrence of at least one of the certain events.

### Claim 14 (depends on 13)

A tangible, non-transitory computer readable medium having instructions thereon which, upon execution by one or more processors, alone or in combination, provide for execution of the method according to claim 13.

### Claim 15 (depends on 1)

A tangible, non-transitory computer readable medium having instructions thereon which, upon execution by one or more processors, alone or in combination, provide for execution of the method according to claim 1.

### Claim 16 (depends on 4)

The method according to claim 4, wherein the network comprises a plurality of computing devices used by one or more users, wherein the features of the at least two feature types include one or more device features of the computing devices, one or more user features of the users and one or more request features of requests initiated by the users, and wherein the network is a communications network and the one or more events predicted by the joint representation are security events.
