System and method for multi-objective reinforcement learning

A computer-implemented method for processing multiple input objectives by a reinforcement learning agent, the method comprising:

• instantiating a reinforcement learning agent that maintains a reinforcement learning neural network interacting with an environment and generates, according to outputs of the reinforcement learning neural network, signals for communicating task requests;
• receiving current state data characterizing a state of the environment;
• receiving a plurality of input data representing a plurality of current user objectives associated with a task request, the plurality of current user objectives associated with a current importance defining a relative importance of each of the plurality of current user objectives;
• receiving a set of historical task data associated with historical task requests for the resource, the set of historical task data including historical state data characterizing at least one historical state of the environment, and at least one corresponding historical action of the reinforcement learning agent;
• sampling, from the set of historical task data, a subset of the set of historical task data associated with a plurality of historical user objectives having a historical importance defining a relative importance of each of the plurality of historical user objectives, the historical importance being different from the current importance;
• generating augmented data including the subset of the set of historical task data and the plurality of current user objectives;
• computing a reward based on the augmented data, the historical importance and the current state data;
• updating network parameters of the reinforcement learning neural network based on the reward to train the reinforcement learning agent;
• and
• generating, based on the trained reinforcement learning neural network, the current state data, and the plurality of input data, a signal for communicating the task request.

A non-transitory computer-readable storage medium storing instructions which when executed cause at least one computing device to:

• instantiate a reinforcement learning agent that maintains a reinforcement learning neural network interacting with an environment and generates, according to outputs of the reinforcement learning neural network, signals for communicating task requests;
• receive current state data characterizing a state of the environment;
• receive a plurality of input data representing a plurality of user objectives associated with a task request for a resource, the plurality of current user objectives associated with a current importance defining a relative importance of each of the plurality of current user objectives;
• receive a set of historical task data associated with historical task requests for the resource, the set of historical task data including historical state data characterizing at least one historical state of the environment, and at least one corresponding historical action of the reinforcement learning agent;
• sample, from the set of historical task data, a subset of the set of historical task data associated with a plurality of historical user objectives having a historical importance defining a relative importance of each of the plurality of historical user objectives, the historical importance being different from the current importance;
• generate augmented data including the subset of the set of historical task data and the plurality of current user objectives;
• compute a reward based on the augmented data, the historical importance and the current state data;
• update network parameters of the reinforcement learning neural network based on the reward to train the reinforcement learning agent;
• and
• generate, based on the trained reinforcement learning neural network, the current state data, and the plurality of input data, a signal for communicating the task request.