When Can You Correct Distribution Drift in Temporal Graph Generation? A Sharpening--Drift Tension and an Impossibility for Observation-Based Correction

2026-07-27

AI Summary

This study addresses the severe performance degradation of temporal graph generation models under distribution shift and their limited recoverability via observed data. For the first time, it rigorously establishes a theoretical limit on observation-based correction methods, demonstrating that the unpredictability of drift trends and the mean-reverting nature of the drift process fundamentally invalidate extrapolation. Through masked flow matching loss decomposition, lower-bound analysis of conditional variance, a measurability criterion for trend extrapolation, and large-scale sampling experiments, the work quantifies the intrinsic tension between model sharpening and distribution shift. Empirical results show that distribution shift inflates the error floor by 2.2–34.3×; even the best observation-based correction recovers only 5.7% of oracle performance, while extrapolation performs worse than no correction at all.

Abstract

Generative models of temporal graphs are trained on one stretch of an evolving network and deployed on the next, and they degrade badly in the gap. We show this degradation is derivable, general, and not fixable from observations. The masked flow-matching loss decomposes exactly, with no independence assumption, into an irreducible entropy plus a divergence whose derivative along the training path is positive precisely for structures rare during training and common at deployment, diverging as their training probability goes to zero. Empirically the trade-off is a power law with exponent $-0.605$ ($R^2=0.9977$), and drift raises the sampler's error floor without changing how many steps reach it: across seven well-powered conditions the drift-period marginal error varies by at most $6\%$ over a $50\times$ range of sampling budgets, while the floor sits $2.2\times$ to $34.3\times$ above the in-period floor. Because the deployment period is observed, correction looks like a matter of measurement. It is not. We prove that any corrector measurable with respect to past observations leaves at least the conditional variance of the statistic it tracks, and that trend extrapolation beats trusting the last observation only when $μ^2>v(1-2ρ)$. Both premises are measurable and both go the wrong way: the drift is trendless and mean-reverting, with a one-step innovation as large as the drift itself. An oracle removes $60\%$ of the error, the best observation-based corrector recovers $5.7\%$ of that, and extrapolation is strictly worse than doing nothing clever.

Problem

• distribution drift
• temporal graph generation
• observation-based correction
• generative models
• model degradation

Innovation

• distribution drift
• temporal graph generation
• masked flow-matching
• impossibility result
• observation-based correction