Distinguish Permanent Flow Changes from Sampling Noise

Develop a change-detection test for the Kairos mixing-weight and presence predictors that distinguishes a permanent change in a location’s pedestrian flow from sampling noise and resets the affected cell rather than progressively diluting the change through cumulative averaging.

Background

Kairos estimates mixing weights and presence rates using cumulative means, so each new observation receives a weight that decreases as the number of accumulated observations grows. This makes the estimator stable under recurring patterns but slow to adapt when pedestrian flow changes permanently. The paper explicitly states that no existing test in the presented system can distinguish such a permanent change from ordinary sampling noise. The proposed direction is to add recency weighting and a change test that resets a cell when a genuine change is detected.

References

First, the mixing-weight and presence predictors accumulate cumulative means, so every observation carries equal weight: a place whose flow changes permanently is corrected at a rate of one over its accumulated sample count, and no test separates such a change from sampling noise.

— Kairos: Grounded Forecasting of Presence and Directional Flow in 4D Scene Graphs  (2609.27467 - Catalano et al., 23 Sep 2026) in Section 8, Conclusion, subsection “Limitations” and “Future Work”