Nvidia patents method to balance data types when training neural networks
Nvidia describes a technique for training artificial intelligence models that adjusts how much the system cares about mistakes in different categories. The method changes the importance of each category during training based on how much data is available for it, helping the model learn from uneven datasets.
In the filing’s own words
Apparatuses, systems, and techniques are presented to train neural networks and use those neural networks for inferencing tasks. In at least one embodiment, one or more neural networks are caused to be trained using weight parameters based, at least in part, on an amount of training data used to train the one or more neural networks.
What exactly they claim (claim 1)
1. A processor, comprising: one or more circuits to cause one or more neural networks to be trained to classify data according to a plurality of classes, wherein to cause the one or more neural networks to be trained the one or more circuits are further to: calculate respective losses according to a loss function for different classes of the plurality of classes based on inferences of the different classes by the one or more neural networks, wherein calculation of the respective losses comprises adjusting a respective loss for each of at least two of the different classes using a different class-specific weight value parameter for each of the at least two different classes during training; cause network parameters of the one or more neural networks to be updated during the training of the one or more neural networks based on the respective losses as adjusted using the different class-specific weight values; and update one or more of the class-specific weight values for subsequent training based, at least in part, on an amount of loss according to the loss function for a corresponding class of the at least two different classes during the training.
Facts
- Company
- Nvidia (NVIDIA Corporation)
- Official title
- Dynamic class weighting for training one or more neural networks
- Patent number
- US12743876B2
- Application number
- 18203562
- Filed
- May 30, 2023
- Granted
- Sep 22, 2026
- Inventors
- 3
- Claims
- 30
Classified under
G06V 10/82 Physics & computingMore from Nvidia in AI & software
- Nvidia patents AI coach that analyzes your gameplay to suggest strategies Oct 13, 2026
- Nvidia patents AI system that writes and voices character conversations Sep 10, 2026
- Nvidia patents system that uses AI to find bugs in game videos Sep 17, 2026
- Nvidia patents AI method to make digital characters move realistically Sep 8, 2026
- Nvidia patents AI system that builds image generators from text prompts Sep 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.