Causal Future Prediction in a Minkowski Space-Time

Athanasios Vlontzos, Henrique Bergallo Rocha, Daniel Rueckert, Bernhard Kainz

Abstract

Estimating future events is a difficult task. Unlike humans, machine learning ap- proaches are not regularized by a natural understanding of physics. In the wild, a plausible succession of events is governed by the rules of causality, which can- not easily be derived from a finite training set. In this paper we propose a novel theoretical framework to perform causal future prediction by embedding spatio- temporal information on a Minkowski space-time. We utilize the concept of a light cone from special relativity to restrict and traverse the latent space of an arbitrary model. We demonstrate successful applications in causal image synthe- sis and future video frame prediction on a dataset of images. Our framework is architecture- and task-independent and comes with strong theoretical guarantees of causal capabilities.
BibTeX
@misc{vlontzos2020causal,
  author = {Athanasios Vlontzos and Henrique Bergallo Rocha and Daniel Rueckert and Bernhard Kainz},
  title = {{Causal Future Prediction in a Minkowski Space-Time}},
  year = {2020},
  eprint = {2008.09154},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2008.09154},
}
Figure