Samsung Patents a System to Track and Manage AI Models Running Phone Networks
Before a phone network finds out that a new piece of equipment is failing, it usually waits for an alarm to go off. Samsung's patent describes a way to guess in advance by comparing the newcomer with older gear that already broke.

- Publication number
- US 2026/0370557 A1(US20260370557A1)
- Applicant
- SAMSUNG ELECTRONICS CO., LTD.
- Filed · Published
- Oct 15, 2025 · Nov 19, 2026
- Inventors
- Chadi KHIRALLAH, Mutasem HAMDAN, Ezeddin Al HAKIM
- Status when we published
- Waiting for an examiner
- Document
- 15 claims
- US classification
- 709/223
- Parent application
- National Stage Entry of PCT/KR2024/005018 (filed Apr 15, 2024)(PCTKR2024005018)
What Samsung's network model registry does
Phone networks today mostly learn about a problem after an alarm goes off. The fault systems work from reported alarms and don't try to guess what will go wrong next. Samsung's patent describes a way to change that as networks fill up with more equipment and more AI software.
The idea is to turn each piece of equipment into a short list of numbers, a kind of fingerprint built from details like its maker, its alarms, its location and even the weather. Equipment with similar fingerprints is likely to have similar troubles. So when a company installs a new radio unit, the network can compare it with older ones and flag it if it looks like a unit that already overheated.
To keep track of all the AI models that produce those fingerprints, the patent adds a dedicated manager. It registers the models, hands them out on request, and updates them when the network changes.
transmitting, to a second network entity/function, a first message relating to a representation-model management procedure; and as a response to the first message, receiving, from a second network entity, a second message relating to the representation-model management procedure.
Translation: One network part sends an instruction about managing an AI model and waits for a reply.
How a new radio unit gets compared to old ones
At its broadest, the filing covers a simple exchange: one piece of network equipment sends a message about managing a representation model, and another sends a reply. A representation model is an AI model that turns a messy description of something, like a radio unit's maker, alarms and antenna count, into a short list of numbers (a vector).
The patent adds a Representation-Model Management Controller (RMC), a dedicated function that can sit in the radio network, the core network, an edge server or the cloud. Clients register models with it through a request and a response. Each model gets a profile listing its ID, version, size, type, accuracy score and where it was trained.
The controller also watches for changes. When new equipment or software appears, it can trigger registering, configuring or training a model for it, and it can combine models from different parts of the network into one.
In the worked example, an auto-encoder (a neural network that squeezes data into a short code) shrinks six facts about each radio unit to two numbers. The system then measures the Euclidean distance (plain straight-line distance) between those numbers. If a new unit sits close to one that already has a high-temperature alarm, the system predicts the new one will get the same alarm. The same trick is used to predict clashes between automated network-tuning programs.
The present disclosure relates to a 5G communication system or a 6G communication system for supporting higher data rates beyond a 4G communication system such as long term evolution (LTE).
Translation: The technology applies to advanced 5G and 6G mobile networks that handle much heavier data traffic.
Why 6G networks need a librarian for AI models
For you, the payoff would be a phone network that spots trouble before it spreads. If a new tower component looks like one that already failed, an operator could check it early instead of waiting for dropped calls. The filing also targets clashes where two automated tuning programs fight over the same settings, such as antenna tilt and transmit power.
The bigger point is bookkeeping. The filing says 6G networks, expected around 2030, will connect so many devices and AI models that someone has to track which model belongs to which equipment, which version is current and how well it performs. Samsung is proposing a standard set of messages for that job. The claims stay broad and do not spell out the prediction step itself.
This is the 98th Samsung filing we've tracked since May in our 5G and network work, following one on storing AI models and one on blocking AI apps by location.
Source: US 2026/0370557 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.
Get our take in your Top Stories
Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Samsung patent news breaks.
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings · 6 sheets
6 drawing sheets from US 2026/0370557 A1 · click any drawing to enlarge

No comments yet