Research map / Architectures and efficiency / Attention and transformers
WaveSSM: Multiscale State-Space Models for Non-stationary Signal Attention
Ruben Solozabal, Velibor Bojkovic, Hilal AlQuabeh, Klea Ziu, Kentaro Inui, Martin Takáč
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Legendre polynomial bases with wavelet frames in S4 | S4 / WaveSSM · PTB-XL overall AUROC | AUROC: 0.934 -> 0.942 | Improved |
| replace Legendre bases with wavelet frames | S4 / WaveSSM · Speech Commands SC35 raw 16kHz | accuracy (%): 96.08 -> 96.55 | Improved |
| test wavelet SSM under zero-shot downsampling to 8kHz | WaveSSM · Speech Commands SC35 8kHz | accuracy (%): 91.59 (S4 FouT) vs 90.14 (Morlet scaled) | Hurt |
| replace Legendre bases with wavelet frames | S4 / WaveSSM · Long Range Arena (ListOps, Text, Retrieval, Image) | accuracy (%): ListOps 59.60 -> 61.79 | Improved |
| replace Legendre bases with wavelet frames | S4 / WaveSSM · LRA PathX | accuracy (%): 96.35 -> infeasible | Hurt |
| use Gauss, Mexhat or dpss frames with scaled measure (ablation) | WaveSSM · LRA Pathfinder | accuracy (%): 94.20 -> infeasible | Hurt |
| use translated measure instead of scaled measure for Morlet (ablation) | WaveSSM Morlet · LRA Pathfinder | accuracy (%): 94.94 -> 73.3 | Hurt |
| replace Legendre with wavelet frames in S4 | S4 / WaveSSM · Informer univariate forecasting MSE | MSE: ECL 0.432 -> 0.268 | Improved |
| add frame tightening (whitening of the frame) to wavelet frames (ablation) | WaveSSM frames · condition number of frame operator | condition number | Improved |
| use wavelet frames instead of Legendre polynomials for window copying | WaveSSM · Continuous Window Copying | reconstruction MSE | Improved |
| add input-dependent selection (Mamba) instead of linear S4 | S4 · Continuous Window Copying | reconstruction MSE | Improved |
Closest papers by meaning
- Understanding the Differences in Foundation Models: Attention, State Space Models, and Recurrent Neural Networks (NeurIPS 2024 · similarity 0.85)
- Quantifying Memory Utilization with Effective State-Size (ICML 2025 · similarity 0.85)
- TIDES: Implicit Time-Awareness in Selective State Space Models (NeurIPS 2026 · similarity 0.85)
- MaRK: Markov-adapted Recurrent Kernels for Dynamic Operator Conditioning in State Space Models (NeurIPS 2026 · similarity 0.85)
- Autocorrelation Matters: Understanding the Role of Initialization Schemes for State Space Models (ICLR 2025 · similarity 0.85)
- Wavelet-based Positional Representation for Long Context (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
