Disentangled Generative Models for Robust Prediction of System Dynamics
This addresses robustness issues in dynamical system prediction for applications like video analysis, but it is incremental as it builds on existing disentanglement and causal methods.
The paper tackled the problem of improving out-of-distribution generalization and long-term stability in deep neural networks for dynamical system prediction by disentangling domain parameters from dynamics in generative models, resulting in better adaptation to unseen domain parameters and improved predictions in video sequences.
Deep neural networks have become increasingly of interest in dynamical system prediction, but out-of-distribution generalization and long-term stability still remains challenging. In this work, we treat the domain parameters of dynamical systems as factors of variation of the data generating process. By leveraging ideas from supervised disentanglement and causal factorization, we aim to separate the domain parameters from the dynamics in the latent space of generative models. In our experiments we model dynamics both in phase space and in video sequences and conduct rigorous OOD evaluations. Results indicate that disentangled VAEs adapt better to domain parameters spaces that were not present in the training data. At the same time, disentanglement can improve the long-term and out-of-distribution predictions of state-of-the-art models in video sequences.