Segmenting hybrid trajectories using latent ODEs Conference Paper
| Authors: | Shi, R.; Morris, Q. |
| Title: | Segmenting hybrid trajectories using latent ODEs |
| Conference Title: | 38th International Conference on Machine Learning (ICML) |
| Abstract: | Smooth dynamics interrupted by discontinuities are known as hybrid systems and arise commonly in nature. Latent ODEs allow for powerful representation of irregularly sampled time series but are not designed to capture trajectories arising from hybrid systems. Here, we propose the Latent Segmented ODE (LatSegODE), which uses Latent ODEs to perform reconstruction and changepoint detection within hybrid trajectories featuring jump discontinuities and switching dynamical modes. Where it is possible to train a Latent ODE on the smooth dynamical fows between discontinuities, we apply the pruned exact linear time (PELT) algorithm to detect changepoints where latent dynamics restart, thereby maximizing the joint probability of a piece-wise continuous latent dynamical representation. We propose usage of the marginal likelihood as a score function for PELT, circumventing the need for model-complexity-based penalization. The LatSegODE outperforms baselines in reconstructive and segmentation tasks including synthetic data sets of sine waves, Lotka Volterra dynamics, and UCI Character Trajectories. |
| Journal Title | Proceedings of Machine Learning Research |
| Volume: | 139 |
| Conference Dates: | 2021 Jul 18-24 |
| Conference Location: | Virtual |
| ISBN: | 2640-3498 |
| Publisher: | Journal Machine Learning Research |
| Date Published: | 2021-01-01 |
| Start Page: | 9569 |
| End Page: | 9579 |
| Language: | English |
| ACCESSION: | WOS:000768182705066 |
| PROVIDER: | wos |
| Notes: | Proceedings Paper -- Source: Wos |
MSK Authors
-
36Morris
Related MSK Work