US 12,393,598 B2
Automatic slice discovery and slice tuning for data mining in autonomous systems
Nicholas Bien, New York, NY (US); Yunjie Zhao, San Francisco, CA (US); Matthew Elkherj, Belmont, CA (US); Pratik Prabhanjan Brahma, Santa Clara, CA (US); Zehao Hu, Redwood City, CA (US); Or Cohen, San Francisco, CA (US); and Jason Lwin, San Francisco, CA (US)
Assigned to GM CRUISE HOLDINGS LLC, San Francisco, CA (US)
Filed by GM CRUISE HOLDINGS LLC, San Francisco, CA (US)
Filed on Jan. 3, 2024, as Appl. No. 18/403,210.
Prior Publication US 2025/0217375 A1, Jul. 3, 2025
Int. Cl. G06F 16/24 (2019.01); G06F 16/2458 (2019.01)
CPC G06F 16/2465 (2019.01) 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
providing, by a processing device hosting a slice discovery machine learning (ML) model, input data to the slice discovery ML model, the input data including a first set of data corresponding to performance data of an autonomous vehicle (AV) and a second set of data corresponding to a scene in which the AV is operating;
identifying, by the slice discovery ML model, attributes of the AV and corresponding thresholds for the attributes that define a slice comprising a collection of data sharing common characteristics;
providing, by the slice discovery ML model, the attributes and the corresponding thresholds defining the slice to a slice miner to mine training data corresponding to the slice for a tailored dataset;
wherein the slice discovery ML model includes an event-based automatic slice discovery model and a model performance-based automatic slice discovery model;
wherein identifying, by the slice discovery ML model, attributes of the AV and corresponding thresholds for the attributes that define a slice includes receiving an identification of a road event and processing at least the second set of data using the event-based automatic slice discovery model; and
wherein identifying, by the slice discovery ML attributes of the AV and corresponding thresholds for the attributes that define a slice includes processing at least the first set of data using the model performance-based automatic slice discovery model.