US 11,995,566 B2
Machine learning knowledge management based on lifelong boosting in presence of less data
Anil Goyal, Dossenheim (DE); Ammar Shaker, Heidelberg (DE); and Francesco Alesiani, Heidelberg (DE)
Assigned to NEC CORPORATION, Tokyo (JP)
Filed by NEC Corporation, Tokyo (JP)
Filed on Jun. 2, 2020, as Appl. No. 16/889,825.
Prior Publication US 2021/0374566 A1, Dec. 2, 2021
Int. Cl. G06N 5/04 (2023.01); G06N 20/00 (2019.01)
CPC G06N 5/04 (2013.01) [G06N 20/00 (2019.01)] 14 Claims
OG exemplary drawing
 
1. A method for lifelong machine learning using boosting, the method comprising:
receiving a new task and a learning sample for the new task, the learning sample comprising a plurality of examples;
learning a distribution of weights over the learning sample using previously learned classifiers from old tasks, wherein at least one of the old tasks is different from the new task;
learning a set of task-specific classifiers for the new task using a boosting algorithm and the distribution of weights over the learning sample, whereby the distribution of weights over the learning sample is updated using the task-specific classifiers for the new task; and
learning weights over the task-specific classifiers for the old tasks using hard examples from the old tasks, the hard examples being a portion of examples of one or more learning samples for the old tasks that have highest weights wherein the highest weights correspond to the lowest classification accuracy of the task-specific classifiers on the portion of the examples.