Methods, systems, apparatuses, and devices for facilitating secure updating of a machine learning model
Inventors
Adolphe, Eric • Jackson, Elesha Marita • Popov, Matvei
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-12395476-B2
Publication Date
2025-08-19
Expiration Date
Abstract
A system for facilitating secure updating of a machine learning model. The system includes a processing device and a first encryption device. The processing device generates a request and updates the machine learning model using an update. The first encryption device includes a first encryption unit and a first communication unit. The first encryption unit encrypts a native packet corresponding to the request and adds a connectionless header forming a first egressing connectionless datagram, decrypts a second encrypted native packet of an ingressing connectionless datagram to obtain the update. The first communication unit adds a complex header to the first egressing connectionless datagram for forming a first packet for delivery to a second encryption device, receives a second packet comprising the second encrypted native packet and a complex header from the second encryption device, removes the complex header and adds a connectionless header for forming the ingressing connectionless datagram.
Core Innovation
The invention relates to a system for facilitating secure updating of a machine learning model. A processing device generates at least one request for updating at least one machine learning model and updates the at least one machine learning model based on at least one update. The system includes a first encryption device communicatively coupled with the processing device, where the first encryption device encrypts native packets corresponding to the request, adds a connectionless header to form an egressing connectionless datagram, and decrypts an ingressing connectionless datagram to obtain the update.
The system further includes a first communication unit paired with a second communication unit of a second encryption device. The first communication unit adds a complex header to the egressing connectionless datagram for delivery, receives a second packet comprising the second encrypted native packet and a complex header, removes the complex header, and adds a connectionless header to form the ingressing connectionless datagram. The second encryption device processes the paired packets in a complementary manner, including removing the complex header, adding a connectionless header to form an ingressing connectionless datagram, adding a complex header to an egressing connectionless datagram, and the second encryption unit decrypting to obtain the at least one request and encrypting updates to produce encrypted native packets.
In addition to the secure updating, the invention includes anomaly/attack detection using a set of machine learning models. The system uses at least one communication interface configured for receiving at least one signal associated with at least one external device, and the processing device analyzes the at least one signal using a set of machine learning models from a plurality of machine learning models. The processing device determines an attack associated with the first encryption device based on the analyzing, generates an alert for the attack, and generates a performance indicator for each of the plurality of machine learning models based on the determining of the attack, wherein generating the at least one request is further based on the performance indicator for each of the plurality of machine learning models.
Claims Coverage
The partial claim set includes two independent claims that share a common architecture for secure updating of a machine learning model and for determining an attack and producing per-model performance indicators that influence which update requests are generated. The inventive features are concentrated in the paired encryption device workflow using connectionless and complex headers and in the model-based attack, alert, and per-model performance-indicator driven request generation.
Request generation and secure updating of a machine learning model
A system comprising a processing device configured for generating at least one request for updating at least one machine learning model and updating the at least one machine learning model based on at least one update.
Paired encryption devices with connectionless and complex headers
A first encryption device that encrypts a native packet corresponding to the at least one request using an encryption key, adds a connectionless header to form a first egressing connectionless datagram, and decrypts an ingressing connectionless datagram to obtain the at least one update; the first communication unit paired with a second communication unit of a second encryption device, where the first communication unit adds a complex header for delivery, receives a second packet with the complex header and the second encrypted native packet, removes the complex header, and adds a connectionless header to form an ingressing connectionless datagram comprising the second encrypted native packet.
Attack determination, alert generation, and per-model performance indicators driving update requests
At least one communication interface configured for receiving at least one signal associated with at least one external device, where the processing device analyzes the at least one signal using a set of machine learning models from a plurality of machine learning models; determines an attack associated with the first encryption device based on the analyzing; generates an alert for the attack; and generates a performance indicator for each of the plurality of machine learning models based on the determining of the attack, wherein the generating of the at least one request is further based on the performance indicator for each of the plurality of machine learning models.
Second encryption device request/update packet processing
The second encryption device includes a second communication unit configured for receiving the first packet comprising the first encrypted native packet and a complex header, removing the complex header, adding a connectionless header to form an ingressing connectionless datagram comprising the first encrypted native packet, and adding a complex header to an egressing connectionless datagram for forming the second packet for delivery to the first encryption device; and a second encryption unit configured for decrypting the first encrypted native packet of the ingressing connectionless datagram using the encryption key to obtain the at least one request, encrypting the at least one update received from the at least one second device using the encryption key to create the second encrypted native packet, and adding a connectionless header to the second encrypted native packet to form an egressing connectionless datagram.
Across the two independent claims, the claim coverage centers on a secure updating system where encrypted native request and update packets are exchanged between paired encryption devices using connectionless datagrams and complex headers, and where the processing device analyzes external signals with a plurality of machine learning models to determine an attack, generate an alert, and compute per-model performance indicators that further drive the generation of update requests.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?