Learning Data Augmentation Policies
Patent No. EP4462306 (titled "Learning Data Augmentation Policies") was filed by Google on May 20, 2019. The application was issued on Jul 2, 2025.
Patent Summary
A computer-implemented method determines a data augmentation policy for training a machine learning model on image processing tasks (classification or bounding-box regression) or speech recognition tasks. Iteratively, until a termination criterion is met, the method generates a current policy comprising multiple sub-policies. Each sub-policy specifies a sequence of transformation operations, their magnitudes, and application probabilities. The model is trained by transforming input batches according to a randomly selected sub-policy and adjusting model parameters. A quality measure, characterizing model performance on the task, is determined for each policy. Finally, a final data augmentation policy is selected based on these quality measures to train the model, increasing training input quantity and diversity to improve task effectiveness.
Patent Family
Patent Oppositions
Patent oppositions filed by competitors challenge the validity of a granted patent. These oppositions are typically based on claims of prior art, lack of novelty, or non-obviousness. They are a key part of the process for determining a patent's strength and enforceability.
Dossier Documents
The dossier documents provide a comprehensive record of the patent's prosecution history - including filings, correspondence, and decisions made by patent offices - and are crucial for understanding the patent's legal journey and any challenges it may have faced during examination.
Date
Description
Get instant alerts for new documents
EP4462306
- Application Number
- EP24201809A
- Filing Date
- May 20, 2019
- Status
- Granted And Under Opposition
- May 29, 2025
- Publication Date
- Jul 2, 2025
- External Links
- Slate, Register , Google Patents