Utility of Symbolic Priors
Background: Structural priors (e.g., periodicity, trends) can be injected into transformer-based factor graphs to guide learning in data-scarce regimes.
Question / Future Work: The performance benefits of injecting specific symbolic time-series priors into probabilistic factor graphs remain inconclusive, specifically regarding how to systematically evaluate the utility of these priors across diverse multi-pattern datasets without overfitting or degrading performance compared to purely data-driven representations.
Why It Matters: This addresses the fundamental trade-off between domain-informed structural inductive biases and the flexibility of deep learning in time-series forecasting.
Metadata & Links
- created_at
- 2026-05-02T05:49:22Z