A method for training a neural network that automatically selects (classifies) training data (e.g. images, video, audio) based on metadata 204 (e.g. labels) associated with the training data. The training data may be processed into groups based on a combination of metadata values and may be selected based on the groups. The training data may be used in human vision emulation, speech recognition, natural language processing, recommendation systems etc. The automatic selection of the metadata may use an equation solver. The metadata may concern an image captured using a medical device or an autonomous vehicle, including an airplane or robotic vehicle, at a given point in time. The metadata may indicate one or more operational design domain (ODD) values. For example, given a set of labelled scenes and a target distribution of various categorical metadata around a scene, the neural network may classify training images based on sets of the labelled scenes. Other data that may be used concern video, audio, location, weather data etc.
Selecting training data for neural networks
2023-06-07
Patent
Electronic Resource
English
IPC: | G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / B60W CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION , Gemeinsame Steuerung oder Regelung von Fahrzeug-Unteraggregaten verschiedenen Typs oder verschiedener Funktion / G06V |
Selecting and training drivers
Engineering Index Backfile | 1928
Selecting and training transport pilots
Engineering Index Backfile | 1931
|Selecting and training fork truck operators
Engineering Index Backfile | 1952
|Selecting Global Texture Features for Optimum Training
British Library Conference Proceedings
|