analiza aiqbench
The paper tests whether matching a synthetic data source's spectral predictability (Ω) to a target is enough to choose useful pretraining data for time-series forecasting. Across LSTM, PatchTST, and TCN models, Ω-matching helped in controlled synthetic-to-synthetic settings but gave only a marginal average gain on the real ETTm1 benchmark, and chaotic ODE sources degraded performance. The authors find Ω barely predicts which source will transfer (r = -0.02), so practitioners should not treat it as a standalone source-selection criterion.
Is spectral predictability enough? Testing {\Omega}-matching for synthetic pretraining under distribution shift
arXiv:2610.30802v1 Announce Type: new Abstract: Synthetic pretraining under distribution shift requires deciding what synthetic data to use. A natural criterion is to match the spectral predictability $\Omega$ of the source to the target, but is $\Omega$ sufficient? We test this hypothesis, which we call predictability-aware pretraining, across three architectures (LSTM, PatchTST, TCN) and three seeds. Under ideal synthetic-to-synthetic conditions, $\Omega$-matched pretraining can outperform even an oracle baseline (+46.8% MSE reduction, LSTM), though the benefit is already architecture-dependent. On the real-world ETTm1 benchmark, however, this relationship does not hold: sine-plus-noise pretraining at matched $\Omega$ yields only a marginal mean improvement +2.3% averaged over all three architectures, up to +65.5% on individual channels), whereas $\Omega$ matched chaotic ODE trajectories systematically degrade performance -13.2%). The two sources agree on the direction of transfer for only 53% of configurations across all three architectures (Pearson r=-0.02), indicating that $\Omega$ carries little information about which source will transfer beneficially. We further observe an architecture-dependent asymmetry in the response to the direction of the $\Omega$ gap. We conclude that $\Omega$ matching can help in controlled settings but is not sufficient to predict transfer across sources, and we discuss additional signal properties that a source-selection criterion may need to consider.
Title:Is spectral predictability enough? Testing Ω-matching for synthetic pretraining under distribution shift
Abstract:Synthetic pretraining under distribution shift requires deciding what synthetic data to use. A natural criterion is to match the spectral predictability $\Omega$ of the source to the target, but is $\Omega$ sufficient? We test this hypothesis, which we call predictability-aware pretraining, across three architectures (LSTM, PatchTST, TCN) and three seeds. Under ideal synthetic-to-synthetic conditions, $\Omega$-matched pretraining can outperform even an oracle baseline (+46.8% MSE reduction, LSTM), though the benefit is already architecture-dependent. On the real-world ETTm1 benchmark, however, this relationship does not hold: sine-plus-noise pretraining at matched $\Omega$ yields only a marginal mean improvement +2.3% averaged over all three architectures, up to +65.5% on individual channels), whereas $\Omega$ matched chaotic ODE trajectories systematically degrade performance -13.2%). The two sources agree on the direction of transfer for only 53% of configurations across all three architectures (Pearson r=-0.02), indicating that $\Omega$ carries little information about which source will transfer beneficially. We further observe an architecture-dependent asymmetry in the response to the direction of the $\Omega$ gap. We conclude that $\Omega$ matching can help in controlled settings but is not sufficient to predict transfer across sources, and we discuss additional signal properties that a source-selection criterion may need to consider.
Machine Learning (cs.LG)
arXiv:2610.30802 [cs.LG]
(or arXiv:2610.30802v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.30802 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 303 cuvinte
Subiecte conexe
A Distribution-to-Distribution Neural Probabilistic Forecasting Framework for Dynamical Systems
arXiv · 16 oct. 2026
CAFE: Counterfactual Prediction via Fast Posterior Estimation
arXiv · 30 sept. 2026
Reinforcement Learning over Predictive Distributions for LLM Regression
arXiv · 7 oct. 2026
Uncertainty Decomposition for Bayes-Filtered Transformers via Bayesian Predictive Inference
arXiv · 21 iul. 2026