US 12,437,191 B2
Training neural network model based on data point selection
Abhishake Jha, Bangalore (IN); and Shehnaz Mohamed, Bangalore (IN)
Assigned to SONY GROUP CORPORATION, Tokyo (JP)
Filed by SONY GROUP CORPORATION, Tokyo (JP)
Filed on Nov. 13, 2020, as Appl. No. 17/097,958.
Prior Publication US 2022/0156577 A1, May 19, 2022
Int. Cl. G06N 3/08 (2023.01); G06F 17/18 (2006.01); G06F 18/2113 (2023.01); G06F 18/214 (2023.01); G06N 3/045 (2023.01)
CPC G06N 3/08 (2013.01) [G06F 17/18 (2013.01); G06F 18/2113 (2023.01); G06F 18/214 (2023.01); G06N 3/045 (2023.01)] 19 Claims
OG exemplary drawing
 
1. An electronic device, comprising:
a memory configured to store:
a first plurality of training data points; and
a first neural network model trained for a classification task of a real-time application,
wherein the first neural network model is trained with the first plurality of training data points; and
circuitry communicatively coupled to the memory, wherein the circuitry is configured to:
retrieve a first plurality of external data points which are different from the first plurality of training data points on which the first neural network model is trained;
apply the first neural network model on the first plurality of external data points to determine a first plurality of impact scores for each external data point of the first plurality of external data points,
wherein the determined first plurality of impact scores indicate a first amount of contribution of each training data point of the first plurality of training data points of the first neural network model towards prediction of a label of each external data point of the first plurality of external data points;
determine a first impact score for each external data point of the first plurality of external data points based on the determined first plurality of impact scores;
select a first set of external data points from the first plurality of external data points based on the determined first impact score for each external data point of the first set of external data points,
wherein the determined first impact score for each external data point of the first set of external data points is above a first impact score threshold;
update the first plurality of training data points stored in the memory with the selected first set of external data points to generate a second plurality of training data points; and
re-train the first neural network model with the generated second plurality of training data points.