Change Detection in Probability Flow ODE: Online Testing in Diffusion Latent Spaces
2026-08-24 • Machine Learning
Machine Learning
AI summaryⓘ
The authors tackle the problem of detecting when a sequence of data changes its underlying pattern, which is hard when we don’t know the exact data distributions before or after the change. They train a model on the data before the change and transform these data points into a simple reference form (a standard Gaussian). Then, by analyzing how new data points differ from this reference, they can spot when a change happens. Their method works online, does not rely on strict assumptions about data distribution, and can detect complex changes in the data.
sequential change-point detectiondistributional shiftconditional densitydiffusion modelprobability flow ODEMaximum Mean DiscrepancyShiryaev–Roberts procedureU-statisticnonparametric detectiontime-ordered data
Authors
Artem Kraevskiy, Artem Prokhorov
Abstract
A rapidly growing range of sequential data tasks, such as identifying trend reversals in financial markets, auto-segmenting video and audio recordings, detecting changes in movement direction from motion sensors cannot be fully addressed without detection of distributional shifts in time-ordered data. We consider a sequential change-point detection problem where the conditional density switches at an unknown time, yet neither the pre- nor post-change distribution admits a closed-form. Classical likelihood-ratio statistics are inapplicable in this settings. A conditional diffusion model, trained on pre-change-point data with a frozen context encoder, defines a deterministic bijection via the probability flow ODE. Pre-change observations are mapped onto standard Gaussian latent variables. Post-change observations, processed through the same frozen map, deviate from this reference. We employ the Maximum Mean Discrepancy as the test statistic, derive closed-form expressions for its components under the Gaussian null, and establish its asymptotic distribution as a degenerate U-statistic. Afterwards we apply an online detection procedure of Shiryaev--Roberts to the resulting statistic with exact threshold calibration. The method detects arbitrary distributional shifts, including covariance rotations and higher-order structural breaks, without parametric assumptions on either regime.