TBNAI
TIDES: Implicit Time-Awareness in Selective State Space Models
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 04/10/2026, 20:00:00 UTC
Briefing created: 08/11/2026, 00:10:51 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 08/10/2026, 12:18:24 UTC.
Briefings
TIDES reconciles selective state space models by moving input dependence onto the diagonal state matrix, enabling native handling of irregular timestamps while preserving per-token expressivity. The model achieved the new best average rank on UEA time series classification and matched or exceeded the reference baseline on 6 of 8 natively irregular datasets from astronomy, agriculture, and climate events.
Key points
- TIDES moves input dependence off the step size onto the diagonal state matrix, allowing irregular timestamps without sacrificing per-token expressivity.
- TIDES sets the new best average rank on UEA time series classification and matches or exceeds the reference baseline on 6 of 8 natively irregular datasets.
Context
Selective state space models like Mamba achieve strong per-token expressivity by making the time discretization step a learned function of the input, but this limits handling of irregular time series. Continuous time SSMs like S5 handle irregular timestamps natively but remain linear time invariant, limiting per token expressivity. TIDES reconciles these architectures by moving input dependence onto the diagonal state matrix.
Limitations
The study uses a novel Fading Flash experimental benchmark for controlled diagnostics. Results are from large-scale benchmarks on UEA time series classification and Physiome ODE regression. Performance on 6 of 8 natively irregular datasets is reported; specific failure modes for current SOTA architectures are not detailed in the evidence.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 08/11/2026, 00:10:51 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeianw2efe6h7jopmffmz2qtzet4luhj7issroczpw2lxfysr4n4eiu