analiza aiqbench
TinyCast establishes a new efficiency frontier in zero-shot time-series forecasting by replacing learned attention mechanisms with computed periodicity and dilated convolutions, achieving superior accuracy-to-parameter ratios compared to existing neural models. Its architecture enables direct deployment on embedded devices via static INT8 quantization without requiring per-signal fitting or test-data leakage, addressing critical constraints for resource-constrained AI applications.
TinyCast: Probabilistic Zero-Shot Forecasting with Computed Periodicity
arXiv:2608.15767v1 Announce Type: new Abstract: We introduce TinyCast, an attention-free zero-shot forecaster that emits a predictive distribution from 146,505 parameters, on the premise that at this size the periodic structure of a context is worth computing rather than learning. A zero-parameter spectral detector supplies the dominant periods, the context is folded on their phase, and a dilated convolutional encoder and a block-autoregressive quantile decoder model the rest. It is smaller than every zero-shot entry on the GIFT-Eval board whose parameter count can be established. On probabilistic accuracy it defines the size-accuracy frontier. Among zero-shot entries declaring no test-data leakage it is the only one below 1.4M parameters that emits a predictive distribution, and every entry scoring better carries at least that budget. On Chronos-ZS and fev-bench every neural model ahead of it carries at least 28 times its parameters. Because the mixing path is convolutions and matrix multiplications only, it exports to static INT8 and forecasts end to end on an embedded device without per-signal fitting.
Abstract:We introduce TinyCast, an attention-free zero-shot forecaster that emits a predictive distribution from 146,505 parameters, on the premise that at this size the periodic structure of a context is worth computing rather than learning. A zero-parameter spectral detector supplies the dominant periods, the context is folded on their phase, and a dilated convolutional encoder and a block-autoregressive quantile decoder model the rest. It is smaller than every zero-shot entry on the GIFT-Eval board whose parameter count can be established. On probabilistic accuracy it defines the size-accuracy frontier. Among zero-shot entries declaring no test-data leakage it is the only one below 1.4M parameters that emits a predictive distribution, and every entry scoring better carries at least that budget. On Chronos-ZS and fev-bench every neural model ahead of it carries at least 28 times its parameters. Because the mixing path is convolutions and matrix multiplications only, it exports to static INT8 and forecasts end to end on an embedded device without per-signal fitting.
38 pages, 6 figures, 17 tables. Code and weights: this https URL
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2608.15767 [cs.LG]
(or arXiv:2608.15767v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2608.15767 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 · 270 cuvinte
Subiecte conexe
End-to-End Forecasting with TimesFM 2.5: Backtesting, Covariates, Anomaly Detection, and Scalable Colab Deployment
MarkTechPost · 2 aug. 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · 13 oct. 2026
The Entangled Dimensions of Decision Theory
LessWrong · 30 iul. 2026
Beyond Point Forecasts: A Survey on Probabilistic Forecasting for Time Series and Spatiotemporal Data
arXiv · 15 sept. 2026