Research map / Architectures and efficiency / Pruning and efficient tuning
AIRE-Prune: Asymptotic Impulse-Response Energy for State Pruning in State Space Models
Apurba Prasad Padhy, Fernando F. Camacho, Saibal Mukhopadhyay
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 |
|---|---|---|---|
| score states by asymptotic impulse-response energy with prefix normalization instead of LAST | S5 · LRA average | average accuracy drop (pp) / avg pruning: 0.52 -> 0.29 pp drop; 60.8% vs 33% pruning | Improved |
| use AIRE-Prune instead of Global H-inf | S5 · Pathfinder (30% pruned baseline) | accuracy (%): 57.20 -> 95.15 | Improved |
| use AIRE-Prune instead of LAST on Retrieval | S5 · Retrieval | accuracy (%): 90.42 -> 90.11 | Hurt |
| prune ListOps with AIRE-Prune | S5 · ListOps | accuracy (%): 61.48 -> 61.48 at 20% pruning | No clear effect |
| apply AIRE-Prune to S4D | S4D · Text | accuracy (%): 86.40 -> 85.83 at 90% pruned | Mixed |
| apply AIRE-Prune to Mamba | Mamba · Image | accuracy (%): 69.82 -> 68.97 at 80% pruned | Mixed |
| prune S5 states with AIRE-Prune | S5 · LRA Pathfinder | inference speedup: 2.82x speedup, 64.7% fewer parameters | Improved |
| prune states in S5 by AIRE-Prune on Speech Commands | S5 · Speech Commands | accuracy (%): 96.43 -> 96.40 at 45% pruned | No clear effect |
Closest papers by meaning
- Layer-Adaptive State Pruning for Deep State Space Models (NeurIPS 2024 · similarity 0.92)
- Autocorrelation Matters: Understanding the Role of Initialization Schemes for State Space Models (ICLR 2025 · similarity 0.86)
- Probe Pruning: Accelerating LLMs through Dynamic Pruning via Model-Probing (ICLR 2025 · similarity 0.85)
- Structured Sparse Transition Matrices to Enable State Tracking in State-Space Models (NeurIPS 2025 · similarity 0.85)
- Týr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization (NeurIPS 2025 · similarity 0.85)
- Exploring Token Pruning in Vision State Space Models (NeurIPS 2024 · 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.
