Microsoft Patents

See Microsoft's patents in other areas

Advancements In Training Neural Networks

Microsoft filed these patents around Advancements In Training Neural Networks in the last 5 years

# Patent No. Short Description
1. US20190266246A1 Modeling sequences using segmentations to improve output predictions
2. US20230267319A1 Training and operating neural networks using mixed precision floating point formats to reduce memory and computational requirements
3. US11657799B2 Training techniques for recurrent neural network transducers (RNN-T) to improve accuracy and reduce training time
4. US20210174146A1 Improving an object recognition system by suggesting additional training images that would improve the system's performance
5. US11625627B2 A deep learning framework for microclimate prediction that uses a combination of techniques to improve accuracy and adaptability
6. US10460234B2 Securely training deep neural networks using private data from multiple parties without sharing the data
7. US11809909B2 Providing synthetic data as a service (SDaaS) to democratize and democratize training datasets for machine learning
8. US20230196085A1 Training neural networks using quantized floating-point formats to improve performance on low-power hardware like FPGAs
9. US20200265301A1 Incremental training of a machine learning model using unsupervised data
10. US20200210840A1 Training neural networks using quantized floating-point representations to accelerate training and inference
11. US11741362B2 Training neural networks using mixed-precision computations to reduce training time and resources
12. US10679610B2 Eyes-off training of a dictation system using corrections supplied by a user to improve the system's accuracy
13. US20230186094A1 Probabilistic neural network architecture generation using Monte Carlo methods to avoid fully training intermediate architectures or evaluating the complete search graph
14. US11636389B2 Improving machine learning models for categorizing data by identifying and removing inaccurate training samples to improve the accuracy of the model
15. US11586930B2 Conditional teacher-student learning for improving the performance of student models trained using teacher models

Download the full patent report containing all of the 172 patents