5 problems

Let $S_{\alpha}(l)$ denote the tensor complexity at subsystem size $l$, measured by the $\alpha$-Rényi entropy. Consider tensorized recurrent architectures used for chaotic time-se…

Sparse-spoiling conjecture. Adding a measure can only spoil a predictor on sparse steps, so that the average predictive performance is not affected.

Active-information-storage parameter-selection conjecture. Parameters that maximize $\mathcal{A}_\tau$ also maximize forecast accuracy for LMA, for both maps and flows.

No-integration ARIMA performance conjecture. ARIMA models without an integration term perform more poorly on these five signals, which increases the error and thereby spreads out t…

SPI conjecture. Maximizing SPI minimizes forecast accuracy of LMA, for both maps and flows.