Research map / Architectures and efficiency / Attention and transformers
A Theoretical Analysis of Mamba’s Training Dynamics: Filtering Relevant Features for Generalization in State Space Models
Mugunthan Shandirasegaran, Hongkang Li, Songyang Zhang, Meng Wang, Shuai Zhang
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 |
|---|---|---|---|
| add input-dependent gating (ablation) | Mamba block · locality-structured synthetic data | test accuracy | Improved |
| add input-dependent gating (ablation) | Mamba block · majority-voting synthetic data | test accuracy | Improved |
| increase majority-voting gap alpha_r - alpha_c | Mamba, GD · majority-voting synthetic data | epochs to converge | Improved |
| increase separation Delta L of class-relevant tokens | Mamba, GD · locality-structured synthetic data | convergence speed | Hurt |
| vary feature dimension d among 32, 64, 128 (ablation) | Mamba · majority-voting synthetic data | qualitative behavior: d in {32, 64, 128} | No clear effect |
| vary confusion fraction alpha_c among 0.17, 0.20, 0.23 (ablation) | Mamba · majority-voting synthetic data | agreement with theory: alpha_c 0.17, 0.20, 0.23 | No clear effect |
| use Mamba instead of global attention | single-layer model · locality-structured synthetic data | accuracy | Improved |
Closest papers by meaning
- Demystifying the Token Dynamics of Deep Selective State Space Models (ICLR 2025 · similarity 0.91)
- Block-Biased Mamba for Long-Range Sequence Processing (NeurIPS 2025 · similarity 0.90)
- Generalization Error Analysis for Selective State-Space Models Through the Lens of Attention (NeurIPS 2025 · similarity 0.90)
- Theoretical Foundations of Deep Selective State-Space Models (NeurIPS 2024 · similarity 0.90)
- Understanding Input Selectivity in Mamba: Impact on Approximation Power, Memorization, and Associative Recall Capacity (ICML 2025 · similarity 0.89)
- Mamba Can Learn Low-Dimensional Targets In-Context via Test-Time Feature Learning (NeurIPS 2026 · similarity 0.89)
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.
