# US 11,153,332 B2 · Systems and methods for behavioral threat detection

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

- Number: US 11,153,332 B2
- Document ID: US11153332B2
- Status: Granted
- Kind: B2
- Application number: 16/215,251
- Filed: 2018-12-10
- Published: 2020-06-11
- Granted: 2021-10-19
- Earliest claimed: 2018-12-10
- Assignees: Bitdefender IPR Management Ltd.
- Applicants: Bitdefender IPR Management Ltd.
- Inventors: Daniel Dichiu; Stefan Niculae; Elena A. Bosinceanu; Sorina N. Stoian; Andreea Dincu; Andrei A. Apostoae
- Web page: https://patentgrep.com/patent/US11153332B2

## CPC

- G06F 21/552: {involving long-term monitoring or reporting}
- H04L 63/1416: {Event detection, e.g. attack signature detection}
- G06F 21/554: {involving event detection and direct action}
- G06N 5/02: Knowledge representation; Symbolic representation
- H04L 63/1425: {Traffic logging, e.g. anomaly detection}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

In some embodiments, a behavioral computer security system protects clients and networks against threats such as malicious software and intrusion. A set of client profiles is constructed according to a training corpus of events occurring on clients, wherein each client profile represents a subset of protected machines, and each client profile is indicative of a normal or baseline pattern of using the machines assigned to the client respective profile. A client profile may group together machines having a similar event statistic. Following training, events detected on a client are selectively analyzed against a client profile associated with the respective client, to detect anomalous behavior. In some embodiments, individual events are analyzed in the context of other events, using a multi-dimensional event embedding space.

## Claims

### Claim 1

A server computer system comprising at least one hardware processor configured to:
in response to an occurrence of a target event on a target client system, the target event selected from a group consisting of a launch of a process and an attempt to access a file, determine according to the target event whether the target client system comprises malicious software; and
in response to a determination that the target client system comprises malicious software, perform a security action to protect the target client system;
wherein determining whether the target client system comprises malicious software comprises:
determining an event context of the target event, the event context comprising a set of events having occurred on the target client system prior to the target event and another set of events having occurred on the client system following the target event,
selecting a client profile from a plurality of pre-determined client profiles according to the target event, the selected client profile characterizing a baseline behavior of a selected cluster of client systems,
in response to selecting the client profile, employing a behavior model specific
to the selected client profile to determine a plurality of prediction scores according to the event context, each score of the plurality of prediction scores determined for a distinct event type and indicative of a likelihood that an event of the distinct event type would occur in the event context for the selected client profile,
selecting, from the plurality of prediction scores, a score determined for an event type of the target event, and
comparing the selected score to a pre-determined threshold; and
wherein a composition of the selected cluster of client systems is determined by analyzing a collection of events having occurred on a plurality of client systems to determine which of the plurality of client systems show similar behavior.

### Claim 2 (depends on 1)

The computer system of claim 1, wherein determining whether the target client system comprises malicious software comprises:
when the selected score indicates that the target event is likely to occur in the event context, determining that the target client system does not comprise malicious software; and
when the selected score indicates that the target event is not likely to occur in the event context, determining that the target client system comprises malicious software.

### Claim 3 (depends on 1)

The computer system of claim 1, wherein determining whether the target client system comprises malicious software comprises:
determining a predicted event according to the plurality of prediction scores;
comparing the predicted event to the target event; and
in response, determining whether the target client system comprises malicious software according to a result of the comparison.

### Claim 4 (depends on 1)

The computer system of claim 1, wherein:
the behavior model comprises an event encoder configured to determine a set of coordinates indicative of a position of the target event in an event embedding space; and
the hardware processor is further configured to determine whether the target client system comprises malicious software according to the position.

### Claim 5 (depends on 1)

The computer system of claim 1, wherein determining the composition of the selected cluster of client systems comprises:
constructing a plurality of event categories according to whether a selected event of the collection of events occurs in a selected context of other events of the collection of events;
in response to constructing the plurality of event categories, deciding whether each client system belongs to the selected cluster of client systems according to an event profile of the each client system, the event profile determined according to a count of events occurring on the each client system and belonging to a selected event category of the plurality of event categories.

### Claim 6 (depends on 1)

The computer system of claim 1, wherein training the behavior model comprises:
assembling a training sequence of events having a training central event and a training event context, wherein each member of the training sequence is selected from the collection of events according to whether the each member has occurred on a member of the selected cluster of client systems;
employing the behavior model, in response to receiving the training event context of the training sequence, to produce a training prediction indicative of a likelihood that the training central event belongs to the training sequence; and
adjust a set of parameter values of the behavior model according to the training prediction.

### Claim 7 (depends on 1)

The computer system of claim 1, further configured to select the client profile according to whether the target client system is a member of the selected cluster.

### Claim 8 (depends on 1)

The computer system of claim 1, wherein selecting the client profile comprises:
determining a set of coordinates indicating a position of the target client system in a multidimensional embedding space;
in response, selecting a target cluster of the plurality of clusters according to the set of coordinates; and
selecting the client profile according to the target cluster.

### Claim 9 (depends on 8)

The computer system of claim 8, configured to select the target cluster further according to another set of coordinates indicating a position of a member of the target cluster in the embedding space.

### Claim 10

A computer-implemented method comprising:
in response to an occurrence of a target event on a target client system, the target event selected from a group consisting of a launch of a process and an attempt to access a file, employing at least one hardware processor of a computer system in determining according to the target event, whether the target client system comprises malicious software; and
in response to a determination that the target client system comprises malicious software, performing a security action to protect the target client system;
wherein determining whether the target client system comprises malicious software comprises:
determining an event context of the target event, the event context comprising a set of events having occurred on the target client system prior to the target event and another set of events having occurred on the client system following the target event,
selecting a client profile from a plurality of pre-determined client profiles according to the target event, the selected client profile characterizing a baseline behavior of a selected cluster of client systems,
in response to selecting the client profile, employing a behavior model specific to the selected client profile to determine a plurality of prediction scores according to the event context, each score of the plurality of prediction scores determined for a distinct event type and indicative of a likelihood that an event of the distinct event type would occur in the event context for the selected client profile,
selecting, from the plurality of prediction scores, a score determined for an event type of the target event, and
comparing the selected score to a pre-determined threshold; and
wherein a composition of the selected cluster of client systems is determined by analyzing a collection of events having occurred on a plurality of client systems to determine which of the plurality of client systems show similar behavior.

### Claim 11 (depends on 10)

The method of claim 10, wherein determining whether the target client system comprises malicious software comprises:
when the selected score indicates that the target event is likely to occur in the event context, determining that the target client system does not comprise malicious software; and
when the selected score indicates that the target event is not likely to occur in the event context, determining that the target client system comprises malicious software.

### Claim 12 (depends on 10)

The method of claim 10, wherein determining whether the target client system comprises malicious software comprises:
determining a predicted event according to the plurality of prediction scores;
comparing the predicted event to the target event; and
in response, determining whether the target client system comprises malicious software according to a result of the comparison.

### Claim 13 (depends on 10)

The method of claim 10, wherein:
the behavior model comprises an event encoder configured to determine a set of coordinates indicative of a position of the target event in an event embedding space; and
the at least one hardware processor is further configured to determine whether the target client system comprises malicious software according to the position.

### Claim 14 (depends on 10)

The method of claim 10, wherein determining the composition of the selected cluster of client systems comprises:
constructing a plurality of event categories according to whether a selected event of the collection of events occurs in a selected context of other events of the collection of events;
in response to constructing the plurality of event categories, deciding whether each client system belongs to the selected cluster of client systems according to an event profile of the each client system, the event profile determined according to a count of events occurring on the each client system and belonging to a selected event category of the plurality of event categories.

### Claim 15 (depends on 10)

The method of claim 10, wherein training the behavior model comprises:
assembling a training sequence of events having a training central event and a training event context, wherein each member of the training sequence is selected from the collection of events according to whether the each member has occurred on a member of the selected cluster of client systems;
employing the behavior model, in response to receiving the training event context of the training sequence, to produce a training prediction indicative of a likelihood that the training central event belongs to the training sequence; and
adjusting a set of parameters of the behavior model according to the training prediction.

### Claim 16 (depends on 10)

The method of claim 10, further comprising selecting the client profile according to whether the target client system is a member of the selected cluster.

### Claim 17 (depends on 10)

The method of claim 10, wherein selecting the client profile comprises:
determining a set of coordinates indicating a position of the target client system in a multidimensional embedding space;
in response, selecting a target cluster of the plurality of clusters according to the set of coordinates; and
selecting the client profile according to the target cluster.

### Claim 18 (depends on 17)

The method of claim 17, comprising selecting the target cluster further according to another set of coordinates indicating a position of a member of the target cluster in the embedding space.

### Claim 19

A non-transitory computer-readable medium storing instructions which, when executed by at least one hardware processor of a computer system, cause the computer system to:
in response to an occurrence of a target event on a target client system, the target event selected from a group consisting of a launch of a process and an attempt to access a file, determine according to the target event whether the target client system comprises malicious software; and
in response to a determination that the target client system comprises malicious software, perform a security action to protect the target client system;
wherein determining whether the target client system comprises malicious software comprises:
determining an event context of the target event, the event context comprising a set of events having occurred on the target client system prior to the target event and another set of events having occurred on the client system following the target event,
selecting a client profile from a plurality of pre-determined client profiles according to the target event, the selected client profile characterizing a baseline behavior of a selected cluster of client systems,
in response to selecting the client profile, employing a behavior model specific to the client profile to determine a plurality of prediction scores according to the event context, each score of the plurality of prediction scores determined for a distinct event type and indicative of a likelihood that an event of the distinct event type would occur in the event context for the selected client profile,
selecting, from the plurality of prediction scores, a score determined for an event type of the target event, and
comparing the selected score to a predetermined threshold; and
wherein a composition of the selected cluster of client systems is determined by analyzing a collection of events having occurred on a plurality of client systems to determine which of the plurality of client systems show similar behavior.

### Claim 20 (depends on 1)

The computer system of claim 1, wherein the at least one hardware processor is configured to select the client profile further according to a current user of the target client system.

### Claim 21 (depends on 10)

The method of claim 10, wherein the at least one hardware processor is configured to select the client profile further according to a current user of the target client system.

### Claim 22 (depends on 1)

The computer system of claim 1, wherein the event context of the target event comprises an attempt by the target client system to access a selected network location.

### Claim 23 (depends on 10)

The method of claim 10, wherein the event context of the target event comprises an attempt by the target client system to access a selected network location.
