AI RESEARCH
TIDES: Implicit Time-Awareness in Selective State Space Models
arXiv CS.AI
•
ArXi:2605.09742v1 Announce Type: cross Selective state space models (SSMs), such as Mamba, achieve strong per-token expressivity by making the time discretization step $\Tilde{\Delta}$ a learned function of the input. However, in doing so, $\Tilde{\Delta}$ ceases to represent a physical sampling interval, limiting its irregular time series modeling capability. Continuous-time SSMs, such as S5, preserve the physical meaning of $\Tilde{\Delta}$ and handle irregular natively ($\Tilde{\Delta}\equi\Delta)$, but their dynamics remain linear time-invariant (LTI), limiting per-token expressivity.