Machine Learning Hyperparameter Optimization Patents: Who Is Filing and Where the Field Is Still Open

897 published records match this search across the 2015 to 2026 window, using IPC classes G06N and G06N3 combined with terms for hyperparameter optimization, tuning or parameter search. That count includes applications and grants, so the number of records is not the same as the number of enforceable patents. Because publication lags filing by roughly 18 months, the most recent one to two years will still grow as more filings are published.

The assignee ranking covers 100 companies, with the leader holding 26 records and the fifth-placed filer holding 19. The top five combined account for 12.7% of the 897 records in scope, and the top ten account for 19.7%, which is a moderate concentration rather than a monopoly — most of the field sits in a long tail of filers with just one or a handful of records each. That spread means a freedom-to-operate check needs to look well past the largest few names.

It is growing. Filings rose from 76 records in 2021 to 169 in 2024, a +122% increase over that three-year span, and the tracked peak so far is 247 records in 2025. Because patent publication typically lags filing by about 18 months, 2025 and 2026 numbers are undercounts that will revise upward, so there is no evidence in this data of the field slowing.

US11809968B2, granted to Capital One Services, LLC on 2023-11-07, covers a specific mechanism: training a prediction model to estimate whether continuing a hyperparameter-tuning technique will still improve a chosen metric, then stopping that tuning technique once the prediction model clears an accuracy threshold and forecasts no further gain. It does not claim hyperparameter tuning in general. Tuning systems that stop on a fixed budget or a direct statistical check of the validation metric, without a second gating prediction model, sit outside this specific claim.

Beyond the base G06N class that every record in this search carries by definition, G06F (electronic digital data processing) appears on 34.1% of the 897 records, making it the most common secondary class. G06Q (business, commerce and admin processing) follows at 19.6% and G16H (healthcare informatics) at 14.6%, showing that tuning methods are usually claimed together with a specific application rather than in isolation. Because records can carry multiple IPC codes, these shares add up to more than 100% and should not be summed.

India leads receiving offices with 421 filings in this dataset, ahead of the United States at 251, China at 93, WIPO's PCT route at 62, the European Patent Office at 34 and South Korea at 14. That distribution reflects where applications were filed or entered national phase, not necessarily where the underlying research originated, and a single family can appear at more than one office.