Researchers have introduced the exponentially weighted signature (EWS), a novel continuous-time model that extends state-space models (SSMs) by computing iterated integrals of a path with increments weighted by a learnable generator. This model maintains the group-like structure and universality of signatures while enabling parallel scans. At its most basic, the EWS functions as an SSM, and the researchers have demonstrated its equivalence to linear time-invariant SSMs, Mamba channels, and Mamba-2 heads. Empirically, the EWS has shown superior performance on time-series classification tasks, with depth generally improving accuracy, and has matched or surpassed competing SSMs on regression and forecasting with fewer parameters. AI
IMPACT Introduces a new model architecture that may offer improved performance and parameter efficiency for time-series classification and forecasting tasks.
RANK_REASON The cluster contains an academic paper detailing a new model architecture. [lever_c_demoted from research: ic=1 ai=1.0]
- Alexandre Bloch
- arXiv
- Exponentially Weighted Signature
- Hugging Face
- Mamba
- Mamba-2
- State space models
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
