Microsoft patents split AI model that keeps data encrypted
Microsoft describes a way to split a neural network between a user's device and a remote server. The device handles simple math steps, while the server handles complex steps, keeping sensitive data encrypted throughout the process.
In the filing’s own words
The technology described herein is related to a hybrid neural network that divides operations of a neural network layer between a server and a client device. In an aspect, one or more liner operations of a neural network layer are performed on the client, while non-linear operations, such as an activation function, are performed on the server. In an aspect, the technology described herein maintains network security by encrypting portions of the client-side components. The encrypted portions may be learned values, which may also be described as learned parameters. In aspects, homomorphic encryption is used.
What exactly they claim (claim 1)
1. One or more computer storage media comprising computer-executable instructions that when executed by computing device performs a method of operating a neural network in hybrid deployment, the method comprising: receiving, at a partial server-side neural network, an intermediate neuron-output for a first layer, wherein the intermediate neuron-output is generated by a partial client-side neural network, wherein the partial client-side neural network executes only linear operations comprising weighting and bias operations of the first layer to produce the intermediate neuron-output; generating, at the partial server-side neural network, a final neuron-output for the first layer using the intermediate neuron-output as input, wherein the partial server-side neural network executes only non-linear activation operations of the first layer; encrypting the final neuron-output to form an encrypted final neuron-output; and communicating the encrypted final neuron-output to the partial client-side neural network.
Facts
- Company
- Microsoft (Microsoft Technology Licensing, LLC)
- Official title
- Hybrid machine learning model environment with homomorphic encryption
- Patent number
- US12744657B2
- Application number
- 18635279
- Filed
- Apr 15, 2024
- Granted
- Sep 22, 2026
- Inventors
- 3
- Claims
- 20
Classified under
H04L 9/008 Electricity & electronicsMore from Microsoft in AI & software
- Microsoft patents method to check AI answers against verified facts Oct 20, 2026
- Microsoft patents AI that presents PowerPoint slides for you Sep 3, 2026
- Microsoft patents method to check if LLM answers are based on real data Sep 10, 2026
- Microsoft patents system to blend multiple AI agents into one chat Oct 1, 2026
- Microsoft patents system that answers questions using your screen history Oct 15, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.