US 12,169,792 B2
Adaptive multi-agent cooperative computation and inference
Augusto Vega, Mount Vernon, NY (US); Pradip Bose, Yorktown Heights, NY (US); and Alper Buyuktosunoglu, White Plains, NY (US)
Assigned to INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed by International Business Machines Corporation, Armonk, NY (US)
Filed on Jun. 22, 2018, as Appl. No. 16/015,986.
Prior Publication US 2019/0392333 A1, Dec. 26, 2019
Int. Cl. G06N 7/01 (2023.01); G05D 1/00 (2024.01); G06N 20/00 (2019.01); H04L 9/40 (2022.01); H04L 67/12 (2022.01)
CPC G06N 7/01 (2023.01) [G06N 20/00 (2019.01); H04L 63/1458 (2013.01); G05D 1/0088 (2013.01); H04L 67/12 (2013.01)] 13 Claims
OG exemplary drawing
 
13. A computer system comprising one or more processors, one or more computer-readable memories, and one or more computer-readable storage media, and program instructions stored on at least one of the one or more storage media for execution by at least one of the one or more processors via at least one of the one or more memories, the stored program instructions when executed by a processor causing operations comprising:
receiving, in a camera in a first autonomous vehicle (AV), a first image of a scene, the first image depicting a first view of the scene in a first atmospheric condition;
computing, at the first AV using a local classification model processing the first image, a local classification and a local classification confidence score corresponding to the first image, wherein the local classification model is trained using a first machine learning training protocol;
receiving, at the first AV in response to a broadcast request, a remote classification and a remote classification confidence score, the remote classification and the remote classification confidence score being computed at a second AV using a first remote classification model, wherein the first remote classification model is trained using a different machine learning training protocol from the first training protocol, the remote classification performed on a second image of the scene using a camera in the second AV, the second image of the scene depicting a second view of the scene in a second atmospheric condition;
forming, at the first AV, a consensus classification, the consensus classification comprising a most frequent classification from a set of classifications, wherein the set of classifications comprises all received remote classification, wherein the set of classifications further includes the local classification responsive to determining that the local classification confidence score is above a local confidence threshold;
computing, at the first AV, a consensus classification confidence score corresponding to the consensus classification, the consensus classification confidence score comprising a weighted average of classification confidence scores corresponding to classifications in the consensus classification, a weight of each classification confidence score adjusted according to correctness of a previous classification performed by another device of a same type as a device performing a classification in the consensus classification;
updating, at the first AV, responsive to determining that there are greater than a first threshold number of classifications in the most frequent classification and that the consensus classification confidence score exceeds a second threshold, the local classification model, the updating resulting in an updated local classification model usable in a new scene in the first atmospheric condition;
assigning, at the first AV, responsive to determining that there are equal to or less than the first threshold number of classifications in the most frequent classification or that the consensus classification confidence score is equal to or less than the second threshold, the first view of the scene to the local classification; and
stopping the first AV at a location, the stopping performed responsive to assigning the first view of the scene to the local classification.