IMPROVED MACHINE LEARNING FOR TECHNICAL SYSTEMS

A system for supporting machine learning having one or more processors; a memory configured to store one or more parameters of an estimator; a differentiable estimator of discounted reward, referred to as a value estimator, configured to receive an observation including information representative of one or more observations of an environment and/or system, access parameters of the value estimator from memory and generate a discounted reward estimate, referred to as a value estimate. The system includes one or more eligibility trace generators configured to receive the value estimate; receive from the value estimator a partial derivative of the value estimate with respect to one of the parameters of a value estimation process; receive a signal from the eligibility trace generator and generate an updated signal, referred to as an eligibility trace. The system having one or more bootstrapped gradient estimators configured to receive a reward signal from a technical reward system; receive the value estimate and the eligibility trace; and create a gradient estimate based on an input including: the reward; the eligibility trace; the value estimate; and the gradient of the value estimate.