Estimation--Prediction Tradeoff in Causal Probabilistic Temporal Graphs
2026-06-26 • Machine Learning
Machine LearningInformation TheoryMultiagent SystemsSocial and Information Networks
AI summaryⓘ
The authors explain that when predicting future connections in changing networks, just looking at how accurate predictions are can mix up mistakes from the model with randomness that can’t be avoided. They show that in some cases, getting better at understanding the model’s parameters actually makes making exact predictions harder because of this randomness. They introduce a new way to create test networks where the ground truth is known, letting them check both how well a model predicts future links and how well it understands the underlying causes. Their work suggests that evaluating models should separate errors that can be fixed from uncertainty that is part of the system.
temporal link predictionprobabilistic temporal graphsFisher informationentropybinary logistic modelscausal inferenceCramér-Rao boundparameter estimationirreducible uncertaintypredictive accuracy
Authors
Aniq Ur Rahman
Abstract
Temporal link prediction is usually evaluated by predictive performance on unseen edges, but in probabilistic temporal graphs this criterion can conflate model error with irreducible uncertainty. We study this issue by characterising an inherent estimation--prediction tradeoff in binary logistic models where regimes that maximise Fisher information and improve parameter recoverability are also those with the highest entropy, making individual predictions intrinsically harder even under perfect parameter recovery. We propose a probabilistic causal framework for generating temporal graphs with transient edges and known ground-truth causal structure, allowing temporal link prediction to be evaluated jointly with causal parameter recovery. For the proposed binary logistic parametrisation, we derive the Cramér--Rao bound and validate the tradeoff between parameter estimation error and irreducible predictive loss. Our results show that predictive accuracy alone may not reflect whether a model has learned the underlying causal mechanism, motivating benchmarks that distinguish reducible model error from intrinsic process uncertainty.