Research map / Learning theory and algorithms / Prediction and decision losses
A Control-Theoretic View of Mamba on Stability and Robustness
Liang Cao, Weide Liu, Zhuo Chen, Yan Qin
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 |
|---|---|---|---|
| increase state dimension n from 8 to 64 (ablation) | selective SSM certificate · synthetic SSM | output-error bound: 4.43 -> 38.94 | Hurt |
| increase stability margin alpha from 0.3 to 2.0 (ablation) | selective SSM certificate · synthetic SSM | Kforce: 6.30 -> 1.10 | Improved |
| widen step-size range from [0.1, 0.5] to [0.1, 2.0] (ablation) | selective SSM certificate · synthetic SSM | Kwarp: 0.270 -> 1.284 | Hurt |
| freeze step size (delta_t constant) (ablation) | selective SSM certificate · synthetic SSM | Kwarp: 0.608 -> 0 | Improved |
| widen step-size range (Frozen to Extreme) (ablation) | selective SSM · synthetic SSM, random perturbations | measured output error: approx 0.01 across configurations | No clear effect |
| compare certified linear-in-T bound with measured sensitivity | Theorem 3.3 bound · synthetic SSM, T up to 200 | sensitivity vs horizon: measured slope 5.6e-5 vs Kwarp 0.608 | No clear effect |
| certified robustness bound versus measured worst error | Theorem 3.3 bound · synthetic SSM, T=150 | bound vs actual error: bound 9.51 vs actual 0.021 | Mixed |
| use adaptive step allocation instead of uniform grid (ablation) | stiff-system sampling · stiff ODE, 273 nodes | peak reconstruction error: 9.5e-2 -> 5.9e-3 | Improved |
| apply CQLF SDP bisection to polytope of non-symmetric matrices | Algorithm 1 · polytopic LPV, K=4 | contraction rate: beta* = 0.732, max observed 0.623 | Improved |
| test pair of stable matrices under switching | Algorithm 1 · nilpotent pair A1, A2=A1^T | feasibility | No clear effect |
Closest papers by meaning
- Stabilizing LTI Systems under Partial Observability: Sample Complexity and Fundamental Limits (NeurIPS 2025 · similarity 0.84)
- Demystifying the Token Dynamics of Deep Selective State Space Models (ICLR 2025 · similarity 0.82)
- Safe and Stable Control via Lyapunov-Guided Diffusion Models (NeurIPS 2025 · similarity 0.82)
- TIDES: Implicit Time-Awareness in Selective State Space Models (NeurIPS 2026 · similarity 0.81)
- Stability Analysis of Switched Linear Systems with Neural Lyapunov Functions (AAAI 2024 · similarity 0.81)
- A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems (ICML 2025 · similarity 0.81)
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.
