Accuracy prior and diversity prior based future prediction

A computer-implemented method for accuracy prior and diversity prior based future prediction, comprising:

• generating an accuracy latent representation based on a feature extracted history portion of a time series of information;
• generating a diversity latent representation based on the feature extracted history portion;
• generating an accuracy loss based on the accuracy latent representation;
• generating a preliminary prediction based on the diversity latent representation;
• generating one or more samples based on the preliminary prediction;
• generating a diversity loss based on the preliminary prediction, the accuracy loss, and one or more of the samples;
• and
• training and updating of a diversity prior based on the diversity loss.

A system for accuracy prior and diversity prior based future prediction, comprising:

• an accuracy prior generating an accuracy latent representation based on a feature extracted history portion of a time series of information;
• a diversity prior generating a diversity latent representation based on the feature extracted history portion;
• a decoder:
• generating a first output based on a first decoder input including the accuracy latent representation;
• and
• generating a second output based on a second, diversity decoder input including the diversity latent representation;
• and
• a processor generating an accuracy prior and diversity prior based future prediction based on the first output and the second output,
• wherein the diversity prior is trained during a training stage utilizing the accuracy prior.