IP Library › Granted Patent US 12,665,946
Granted Patent B2
US 12,665,946 · App. 18/893,391 · Granted Jun 23, 2026

Systems and methods for semantic communication for cyber security

Inventors: Saeideh Parsaei Fard (Cupertino, CA); Matthias Sauer (San Jose, CA)
Assignee: Apple Inc.
H04W12/122
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,665,946
App. No.
18/893,391
Filed
Sep 23, 2024
Granted
Jun 23, 2026
Kind
B2
Art Unit
2494
USPC
726/23
Abstract

Disclosed are methods, systems, and computer-readable medium to perform operations including: receiving coded data comprising a plurality of data elements, each data element comprising a coded message; determining, for a coded message and based on a semantic model, a semantic distortion value, the semantic model comprising one or more classifiers configured to identify a desired meaning of a given coded message based on a context of the given coded message; determining a difference between the semantic distortion value and a baseline semantic distortion value; and generating, when the difference exceeds a threshold value, a security message indicating a potential security issue.

Claims (46)

1 . An apparatus comprising one or more processors configured to perform operations for wireless communication, the operations comprising:

receiving coded data comprising a plurality of data elements, the plurality of data elements comprising coded messages;

determining, for a coded message of the coded messages and based on a semantic model associated with the one or more processors, a semantic distortion value, the semantic model comprising one or more classifiers configured to identify a desired meaning of a given coded message based on a context of the given coded message;

determining a difference between the semantic distortion value and a baseline semantic distortion value, wherein determining the difference between the semantic distortion value and the baseline semantic distortion value comprises:

determining, based on the semantic model, a particular semantic distortion value from a baseline message, the particular semantic distortion value being the baseline semantic distortion value; and

determining the difference between the semantic distortion value and the particular semantic distortion value determined from the baseline message; and

generating, when the difference exceeds a threshold value, a security message for transmission, the security message indicating a potential security issue.

2 . The apparatus comprising one or more processors of claim 1 , wherein the semantic distortion value represents a value of a semantic parameter associated with the coded message, and wherein the baseline semantic distortion value represents an expected value for the semantic parameter associated with the coded message.

3 . The apparatus comprising one or more processors of claim 1 , wherein the security message comprises a flag indicating the potential security issue, the flag being transmitted with the coded message to a remote device.

4 . The apparatus comprising one or more processors of claim 1 , wherein the coded message is a first coded message, and wherein determining the difference between the semantic distortion value and the baseline semantic distortion value comprises:

determining, for a second coded message and based on the semantic model, a second semantic distortion value representing the baseline semantic distortion value;

determining a difference between the semantic distortion value and the second semantic distortion value; and

generating, when the difference exceeds the threshold value, the security message indicating the potential security issue.

5 . The apparatus comprising one or more processors of claim 4 , wherein the first coded message and the second coded message are received within a predefined time period during which changes to one or more semantic parameters associated with each of the coded messages should deviate less than predefined thresholds for each of the one or more semantic parameters.

6 . The apparatus comprising one or more processors of claim 4 , wherein the first coded message and the second coded message are received consecutively.

7 . The apparatus comprising one or more processors of claim 1 , wherein

the baseline message is accessed from a local data source.

8 . The apparatus comprising one or more processors of claim 7 , wherein the baseline message comprises a metadata message for content associated with the coded message.

9 . The apparatus comprising one or more processors of claim 1 , wherein determining the difference between the semantic distortion value and the baseline semantic distortion value comprises:

generating an augmented semantic message representing an expected semantic message for a receiving device;

determining, based on the semantic model, a specific semantic distortion value for the augmented semantic message, the specific semantic distortion value being the baseline semantic distortion value;

determining a difference between the semantic distortion value and the specific semantic distortion value; and

generating, when the difference exceeds the threshold value, the security message indicating the potential security issue.

10 . The apparatus comprising one or more processors of claim 1 , wherein determining the difference between the semantic distortion value and the baseline semantic distortion value comprises:

accessing, from a local data source, a red flag message associated with a known security issue for the semantic model;

determining, based on the semantic model, a specific semantic distortion value for the red flag message, the specific semantic distortion value being the baseline semantic distortion value;

determining a difference between the semantic distortion value and the specific semantic distortion value; and

generating, when the difference is within the threshold value, the security message indicating the potential security issue.

11 . A method comprising:

receiving, in a wireless communications session, coded data comprising a plurality of data elements, the data elements comprising a set of first data elements comprising coded messages that are interleaved with a set of second data elements comprising semantic coded messages;

determining, for a coded message of the set of first data elements, a first semantic distortion value, the determining based on a semantic model comprising one or more classifiers configured to identify a desired meaning of a given coded message based on a context of the given coded message;

determining, based on the semantic model, for a semantic coded message of the set of second data elements and that corresponds to the coded message, a second semantic distortion value; and

generating, responsive to determining that a difference between the first semantic distortion value and the second semantic distortion value exceeds a threshold value, a security message indicating a potential security issue.

12 . The method of claim 11 , wherein the semantic model is a local semantic model stored at a local location.

13 . The method of claim 11 , wherein the first semantic distortion value represents a value for a semantic parameter associated with the coded message, and wherein the second semantic distortion value represents an expected value the semantic parameter associated with the semantic coded message.

14 . The method of claim 11 , further comprising:

receiving, with the coded data, an interleaving element that specifies an interleaving period in which at least one of the first data elements is interleaved into the set of second data elements.

15 . The method of claim 14 , wherein the interleaving element specifies a time period or a number of data elements.

16 . The method of claim 14 , wherein the interleaving period includes at least one first data element that includes first information that is redundant with second information of at least one second data element; or

wherein the interleaving period includes at least one first data element that includes first information received instead of second information of at least one second data element.

17 . The method of claim 11 , further comprising:

receiving, in a tracking message with the coded data, a semantic coherence time parameter that specifies a length of time during which changes to one or more semantic parameters of the set of second data elements should deviate less than a specified threshold value.

18 . The method of claim 11 , further comprising:

receiving, with the coded data, a semantic coherence time parameter that specifies a length of time during which differences among semantic distortion values should deviate less than the threshold value unless there is the potential security issue.

19 . The method of claim 18 , wherein the length of time adjusted based on historic data associated with an application, the historic data specifying nominal semantic behavior for the application, anomalous semantic behavior the application, or a combination thereof.

20 . The method of claim 11 , wherein the security message comprises a request to increase an interleave frequency of the set of first data elements with the set of second data elements.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Oct 8, 2024
From: PARSAEI FARD, SAEIDEH; SAUER, MATTHIAS
To: APPLE INC.
Reel/Frame 068830/0244 →
Continuity (4)
Provisional Application 63541915 · Oct 2, 2023
Provisional Application 63541650 · Sep 29, 2023
Provisional Application 63541697 · Sep 29, 2023
Related Publication 20260095771A1 · Apr 2, 2026
References Cited (75)
US 9064210B1 · Hart · 2015 [cited by applicant]
US 10360402B2 · Farkash · 2019 [cited by examiner]
US 10657434B2 · Seow · 2020 [cited by examiner]
US 11514293B2 · Villegas et al. · 2022 [cited by applicant]
US 11563767B1 · Rodriguez · 2023 [cited by applicant]
US 11654934B2 · Huang et al. · 2023 [cited by applicant]
US 11757914B1 · Jakobsson · 2023 [cited by examiner]
US 12518107B2 · Tunstall-Pedoe · 2026 [cited by examiner]
US 20050182765A1 · Liddy · 2005 [cited by applicant]
US 20140283052A1 · Jordan · 2014 [cited by examiner]
US 20150341379A1 · Lefebvre · 2015 [cited by examiner]
US 20200026283A1 · Barnes et al. · 2020 [cited by applicant]
US 20200241546A1 · Sun et al. · 2020 [cited by applicant]
US 20210288852A1 · Jia et al. · 2021 [cited by applicant]
US 20210374352A1 · Li · 2021 [cited by examiner]
US 20220224442A1 · Srivastava et al. · 2022 [cited by applicant]
US 20220311781A1 · Maeda · 2022 [cited by examiner]
US 20230034450A1 · Sridhar et al. · 2023 [cited by applicant]
US 20230186120A1 · Zhao · 2023 [cited by examiner]
US 20230198663A1 · Stocia et al. · 2023 [cited by applicant]
US 20230199746A1 · Vannithamby et al. · 2023 [cited by applicant]
US 20230219561A1 · El-Shaer et al. · 2023 [cited by applicant]
US 20240095346A1 · Cocea · 2024 [cited by examiner]
US 20240146734A1 · Southgate · 2024 [cited by examiner]
US 20250111249A1 · Parsaei Fard et al. · 2025 [cited by applicant]
US 20250112967A1 · Parsaei Fard et al. · 2025 [cited by applicant]
CN 112800247B · 2021 [cited by applicant]
CN 114091673A · 2022 [cited by applicant]
CN 114091673B · 2022 [cited by applicant]
CN 114885370A · 2022 [cited by applicant]
CN 115292726A · 2022 [cited by applicant]
CN 115470799A · 2022 [cited by applicant]
CN 115567877A · 2023 [cited by applicant]
CN 116055512A · 2023 [cited by applicant]
CN 116073889A · 2023 [cited by applicant]
EP 3257244B1 · 2022 [cited by applicant]
WO WO2023054777A1 · 2023 [cited by applicant]
WO WO2023068398A1 · 2023 [cited by applicant]
WO WO2023095932A1 · 2023 [cited by applicant]
WO WO2023113302A1 · 2023 [cited by applicant]
WO WO2024035531A1 · 2024 [cited by applicant]
[No Author Listed], “Expanded 6G vision, use cases and societal values,” D1.2, Hexa-X, 5G PPP, Apr. 30, 2021, 127 pages. [cited by applicant]
Basu et al., “Preserving quality of information by using semantic relationships,” 2012 IEEE International Conference on Pervasive Computing and Communications Workshops, Mar. 19-23, 2012, pp. 58-63. [cited by applicant]
Bosch GmbH, “Discussion on channel access mechanism for sidelink on unlicensed spectrum,” 3GPP TSG RAN WG1 #109-e, R1-2204868, E-Meeting, May 9-20, 2022, 7 pages. [cited by applicant]
Broadcom Corporation, “Further Discussion on LAA DL Multi-channel LBT,” 3GPP TSG-RAN WG1 Meeting #83, R1-157009, Anaheim, CA, USA, Nov. 15-22, 2015, # pages. [cited by applicant]
Cai et al., “Compute- and Data-Intensive Networks: The Key to the Metaverse,” CoRR, Submitted on May 10, 2022, arXiv:2204.02001v2, 8 pages. [cited by applicant]
Choquette-Choo et al., “Label-Only Membership Inference Attacks,” International Conference on Machine Learning, 2021, 139:1964-1974. [cited by applicant]
Cleverhans.io [online], “Privacy and machine learning: two unexpected allies?” Apr. 29, 2018, retrieved on Sep. 26, 2024, retrieved on <http://www.cleverhans.io/privacy/2018/04/29/privacy-and-machine-learning.html>, 12 … [cited by applicant]
Cnet.com [online], “When video-game glitches attack,” Dec. 18, 2012, retrieved on Sep. 26, 2024, retrieved from URL<https://www.cnet.com/tech/gaming/when-video-game-glitches-attack/>, 14 pages. [cited by applicant]
Dasgupta et al., “Machine learning in cybersecurity: a comprehensive survey,” The Journal of Defense Modeling and Simulation: Applications, Methodology, Technology, Sep. 19, 2020, 19(1):57-106. [cited by applicant]
De Cristofaro, “A Critical Overview of Privacy in Machine Learning,” IEEE Security & Privacy, May 18, 2021, 19(4):1-8. [cited by applicant]
Fredrikson et al. “Model Inversion Attacks that Exploit Confidence Information and Basic Countermeasures,” Conference on Computer and Communications Security, Oct. 12, 2015, 12 pages. [cited by applicant]
Goodfellow et al. “Explaining and Harnessing Adversarial Examples,” CoRR, Submitted on Mar. 20, 2015, arXiv:1412.6572v3, 11 pages. [cited by applicant]
Gu et al. “BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chain,” CoRR, Submitted on Mar. 11, 2019, arXiv:1708.06733v2, 13 pages. [cited by applicant]
Tensorflow.org [online], “Federated Learning for Image Classification,” upon information and belief, available no later than Mar. 7, 2019, retrieved on Oct. 15, 2024, retrieved from URL <https://www.tensorflow.org/feder… [cited by applicant]
Huang et al. “Adversarial Machine Learning,” Proceedings of 4th ACM Workshop on Artificial Intelligence and Security, Oct. 2011, pp. 43-58. [cited by applicant]
Kairouz et al., “Advances and Open Problems in Federated Learning,” CoRR, submitted on Mar. 9, 2021, arXiv:1912.04977v3, 121 pages. [cited by applicant]
Kwon et al., “A survey of deep learning-based network anomaly detection,” Cluster Computing, Sep. 27, 2017, 22(1):949-961. [cited by applicant]
Lan et al., “What is Semantic Communication? A View on Conveying Meaning in the Era of Machine Intelligence,” CoRR, Submitted on Oct. 1, 2021, arXiv:2110.00196v1, 40 pages. [cited by applicant]
Letaie et al., “Edge Artificial Intelligence for 6G: Vision, Enabling Technologies, and Applications,” CoRR, Submitted on Nov. 24, 2021, arXiv:2111.12444v1, 33 pages. [cited by applicant]
Li et al., “Multi-Carrier LAA with Adaptive Energy Detection and Carrier Selection,” 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), Honolulu, HI, USA, Sep. 22-25, 2019, 5 pages. [cited by applicant]
Liu et al., “Privacy and Security Issues in Deep Learning: A Survey,” IEEE, Dec. 15, 2020, pp. 4566-4593. [cited by applicant]
Liu et al., “When Machine Learning Meets Privacy: A Survey and Outlook,” CoRR, Submitted on Nov. 24, 2020, arXiv:2011.11819v1, 35 pages. [cited by applicant]
Lyer et al., “A Survey on Semantic Communications for Intelligent Wireless Networks,” CoRR, Submitted on Aug. 10, 2022, arXiv:2202.03705v2, 30 pages. [cited by applicant]
Mireshghallah et al., “Privacy in Deep Learning: A Survey,” CoRR, Submitted Apr. 25, 2010, arXiv:2004.12254v1, 18 pages. [cited by applicant]
Nokia et al., “On Physical Channel Design Framework for SL-U,” 3GPP TSG RAN WG1 #109-e, R1-2203123, e-Meeting, May 9-20, 2022, 9 pages. [cited by applicant]
Papernot et al., “SoK: Security and Privacy in Machine Learning,” 2018 IEEE European Symposium on Security and Privacy, Apr. 24-26, 2018, 399-414. [cited by applicant]
Papernot “A Marauder's Map of Security and Privacy in Machine Learning: An overview of current and future research directions for making machine learning secure and private,” CoRR, Submitted Nov. 3, 2018, arXiv:1811.011… [cited by applicant]
Shao et al., “Task-Oriented Communication for Multi-Device Cooperative Edge Inference,” CoRR, Submitted on Sep. 12, 2023, arXiv:2109.00172v3, 15 pages. [cited by applicant]
Sheatsley et al., “Adversarial Examples for Network Intrusion Detection Systems,” CoRR, Submitted Sep. 9, 2022, arXiv:2011.01183v3, 15 pages. [cited by applicant]
Strinati et al., “6G networks: Beyond Shannon towards semantic and goal-oriented communications,” CoRR, Submitted on Feb. 17, 2021, 52 pages. [cited by applicant]
Tong et al., “A Survey on Intrusion Detection System for Advanced Metering Infrastructure,” Sixth International Conference on Instrumentation & Measurement, Computer, Communication and Control (IMCCC), Jul. 2016, pp. 33… [cited by applicant]
Tramer et al. “Stealing Machine Learning Models via Prediction APIs,” CoRR, Submitted on Oct. 2, 2016, arXiv:1609.02943v2, 19 pages. [cited by applicant]
Wang et al., “Cooperative Car-Following Control: Distributed Algorithm and Impact on Moving Jam Features,” IEEE Transactions on Intelligent Transportation Systems, May 2016, 17(5):1459-1471. [cited by applicant]
Yun et al., “Attention-based Reinforcement Learning for Real-Time UAV Semantic Communication,” CoRR, Submitted on May 22, 2021, arXiv:2105.10716v1, 6 pages. [cited by applicant]