Microsoft patents method to tweak AI models without retraining them
Microsoft describes a way to adjust large language models for specific tasks by adding small, adjustable layers to the existing system. Instead of changing the main model's core settings, this method uses smaller mathematical shortcuts that are easier to update and merge back into the original design.
In the filing’s own words
A computer implemented method obtains neural network-based model base model weight matrices for each of multiple neural network layers. First low-rank factorization matrices are added to corresponding base model weight matrices to form a first domain model. The low-rank factorization matrices are treated as trainable parameters. The first domain model is trained with first domain specific training data without modifying base model weight matrices.
What exactly they claim (claim 1)
1. A computer-implemented method comprising: accessing a transformer-based, general-domain-trained language model having a plurality of layers, each layer comprising a frozen base weight matrix of dimension d; obtaining, for each layer, a pair of low-rank adaptation matrices previously trained on domain-specific data to adapt the language model, the pair having an intermediate dimension r less than d; merging each frozen base weight matrix with a product of its corresponding pair to form an adapted weight matrix; and performing inference on domain input using the adapted weight matrices.
Facts
- Company
- Microsoft (Microsoft Technology Licensing, LLC)
- Official title
- Low-Rank Adaptation of Neural Network Models
- Publication number
- US20260359642A1
- Application number
- 19748900
- Filed
- Jul 21, 2026
- Published
- Nov 12, 2026
- Inventors
- 5
- Claims
- 20
Classified under
G06N 3/084 Physics & computingG06N 3/10 Physics & computingG06N 5/04 Physics & computingMore 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.
