Continuous learning techniques with machine learning systems to adapt to new input data sets

Abstract

Continuous learning may include receiving a first neural network trained using a first training data set to predict outputs; determining whether the first neural network has a successful prediction rate greater than a prediction threshold; and responsive to determining the first neural network does not have a successful prediction rate greater than the prediction threshold, performing processing. The processing may include training the first neural network using a second training data set different than the first training data set; determining that a trigger condition has occurred, wherein the trigger condition includes detecting, during training of the first neural network using the second training data set, that a first weight of the first neural network has a corresponding weight change exceeding a threshold weight change; responsive to determining the trigger condition has occurred, reconfiguring the first neural network; and training the reconfigured first neural network using the second training data set.