US 12,265,901 B2
Method for training and operating an artificial neural network capable of multitasking, artificial neural network capable of multitasking, and device
Dimitrios Bariamis, Hildesheim (DE)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Appl. No. 17/419,940
Filed by Robert Bosch GmbH, Stuttgart (DE)
PCT Filed Jan. 10, 2020, PCT No. PCT/EP2020/050511
§ 371(c)(1), (2) Date Jun. 30, 2021,
PCT Pub. No. WO2020/156780, PCT Pub. Date Aug. 6, 2020.
Claims priority of application No. 102019201188.8 (DE), filed on Jan. 30, 2019.
Prior Publication US 2022/0067482 A1, Mar. 3, 2022
Int. Cl. G06N 3/00 (2023.01); G06N 3/04 (2023.01)
CPC G06N 3/04 (2013.01) 13 Claims
OG exemplary drawing
 
1. A method for training an artificial neural network (ANN) capable of multitasking, the method comprising the following steps:
providing a first path for a first information flow through the ANN, the first path coupling an input layer of the ANN with at least one task-spanning intermediate layer of the ANN, which is common to a plurality of tasks of the ANN that differ from one another, and the first path couples the at least one task-spanning intermediate layer with a respective task-specific ANN segment of each of the plurality of tasks differing from one another;
providing first training data for training task-spanning parameters, which are common to the plurality of tasks of the ANN differing from one another, via the input layer and the first path;
providing at least one task-specific second path for a second information flow through the ANN that differs from the first information flow, the second path (P2) coupling the input layer of the ANN with only a portion of the task-specific ANN segments of the plurality of tasks differing from one another; and
supplying second training data for the training of task-specific parameters via the second path, wherein the at least one task-spanning intermediate layer includes a plurality of task-spanning intermediate layers, and wherein only a last one of the task-spanning intermediate layers in a direction of information flow from the input layer to the task-specific ANN segments branches directly to a task-specific ANN segment belonging to the first path and directly to a task-specific ANN segment belonging to the second path.