Unsupervised learning techniques for temporal difference models

Abstract

Various example implementations relate to unsupervised training of temporal difference models. A temporal difference model can be trained to receive at least a first state representation and a secondstate representation that respectively describe a state of an object at two different times and, in response, output a temporal difference representation that encodes changes in the object between thetwo different times. To train the model, the temporal difference model can be combined with a prediction model that, given the temporal difference representation and the first state representation, seeks to predict or otherwise reconstruct the second state representation. The temporal difference model can be trained on a loss value that represents a difference between the second state representation and the prediction of the second state representation. In such fashion, unlabeled data can be used to train the temporal difference model to provide a temporal difference representation. The present disclosure further provides example uses for such temporal difference models once trained.

Description

technical field [0001] This disclosure relates generally to machine learning. More specifically, the present disclosure relates to unsupervised learning techniques for temporal difference models. Background technique [0002] Machine learning generally refers to the field of computer science that is devoted to enabling machines such as computers to learn without being explicitly programmed. Machine learning involves the study and construction of algorithms or techniques implemented by machines that enable machines to learn from and predict data. In particular, such algorithms can operate by building models from a training set of input observations to represent data-driven predictions or decisions as outputs, rather than strictly following static programming instructions. [0003] A major branch of machine learning techniques includes supervised learning techniques. Supervised learning can involve inferring or learning a function or model from a training data set containin...

AI Technical Summary

Problems solved by technology

Therefore, it may be difficult and / or expensive (if not impossible) to perform supervised learning in certain scenarios

Images