Microsoft patents simple math trick to sort data without heavy AI
Microsoft describes a method for sorting data into categories using a simplified comparison technique. Instead of complex neural networks, it compares new data to a single average example of a group and adjusts the result using a pre-calculated limit. This allows the system to classify items like images or security threats quickly.
In the filing’s own words
A computer-implemented training method includes receiving a plurality of labelled training data items; generating a training data vector from each training data item; and determining prototype vector representative of training data items. A maximum similarity value between the training data vectors and the prototype vector is identified and stored with the prototype vector. A computer-implemented inference method includes generating an input vector from an input data item; calculating a first similarity value between the input vector and the prototype vector; and calculating a second similarity value based on the first similarity value and the maximum similarity value. The method then determines that the input data item is a member of the class of data items based on the second similarity value. The methods provide lightweight classifiers that can be rapidly trained by non-expert end users for applications such as object detection and cybersecurity risk detection.
What exactly they claim (claim 1)
1. A computer-implemented method of classifying an input data item, comprising: receiving a maximum similarity value, wherein the maximum similarity value is representative of a similarity between: a prototype vector representative of a class of data items, and a most similar training data vector of a plurality of training data vectors in the class of data items from which the prototype vector is calculated; receiving an input data item; generating an input vector from the input data item; calculating a first similarity value between the input vector and the prototype vector; calculating a second similarity value based on the first similarity value and the maximum similarity value; and determining that the input data item is a member of the class of data items by applying a threshold to the second similarity value.
Facts
- Company
- Microsoft (Microsoft Technology Licensing, LLC)
- Official title
- CLASSIFYING DATA ITEMS USING REGULARISED SIMILARITY VALUES
- Publication number
- US20260288949A1
- Application number
- 19472423
- Filed
- May 16, 2024
- Published
- Sep 24, 2026
- Inventors
- 3
- Claims
- 14
Classified under
G06F 21/554 Physics & computingG06F 2221/033 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.