← News

ResearchSources: arXiv · 20 Nov 2026

NOGI: continuous-time neural ODE framework infers interaction graphs from irregular time series

Researchers propose NOGI, a neural ODE-based method that learns latent interaction graphs from multivariate time series, decoupling structure inference from fixed observation schedules.

In plain words

Many systems, like brain networks, can be modeled as parts interacting over time. NOGI is a machine learning method that infers which parts influence which others from time-series data, even when measurements are sparse or irregularly spaced. It does this by modeling the system's evolution as continuous in time rather than tied to fixed sampling intervals.

What was announced

The authors argue that most existing neural relational inference methods model time discretely, which they say degrades under sparse or irregular observations and can entangle multi-hop dependencies into spurious direct edges. NOGI instead couples a deterministic graph learner with an MPNN-based Neural ODE, trained end-to-end via the adjoint method, so gradient signals are integrated continuously between observations for both dynamics and structure learning.

In the reported evaluation, NOGI achieved the best structure recovery in 30 of 32 cases across four simulated dynamical systems at two scales, with competitive forecasting and robustness under observation degradation. On the Netsim fMRI benchmark, it reached an average AUPRC of 0.90 and AUROC of 0.96. These figures come from the paper's abstract alone; the full article was not reviewed, and no independent replication is cited.

What it does not show

  • Evidence is limited to the paper's abstract; the full methodology, experimental setup, and peer-review status were not available for verification.
  • Reported benchmark scores are the authors' own claims and have not been independently confirmed.

Sources

Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 20 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.