Research map / Architectures and efficiency / Attention and transformers
Graph Signal Processing Meets Mamba2: Adaptive Filter Bank via Delta Modulation
Yehjin Shin, Seojin Kim, Noseong Park
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 load balance loss (ablation) | HADES · LAMBADA | perplexity: 26.84 -> 21.74 | Improved |
| add diversity loss (ablation) | HADES · LAMBADA | perplexity: 27.40 -> 21.74 | Improved |
| add expert filters on top of shared filters (ablation) | HADES · LAMBADA | perplexity: 27.64 -> 21.74 | Improved |
| add shared filters on top of expert filters (ablation) | HADES · LAMBADA | perplexity: 30.12 -> 21.74 | Improved |
| use spectral-residual routing instead of random selection (ablation) | HADES · LAMBADA | perplexity: 32.77 -> 21.74 | Improved |
| use spectral-residual routing instead of fixed filter set (ablation) | HADES · LAMBADA | perplexity: 27.64 -> 21.74 | Improved |
| add spectral bias on delta vs no bias (ablation) | HADES · LAMBADA | perplexity: 28.79 -> 21.74 | Improved |
| use spectral bias instead of positional bias (ablation) | HADES · WikiText | perplexity: 30.23 -> 31.51 | Mixed |
| use HADES instead of Mamba2 with 58.9% of parameters | HADES · LAMBADA | perplexity: 24.38 -> 21.74 | Improved |
| use HADES instead of Mamba2 | HADES · WikiText | perplexity: 31.34 -> 31.48 | No clear effect |
| scale to 1B HADES vs 1.3B baselines | HADES · zero-shot commonsense (8 tasks avg) | accuracy (%): 52.46 -> 52.92 | Improved |
Closest papers by meaning
- Block-Biased Mamba for Long-Range Sequence Processing (NeurIPS 2025 · similarity 0.86)
- Achilles' Heel of Mamba: Essential difficulties of the Mamba architecture demonstrated by synthetic data (NeurIPS 2025 · similarity 0.85)
- LongMamba: Enhancing Mamba's Long-Context Capabilities via Training-Free Receptive Field Enlargement (ICLR 2025 · similarity 0.85)
- Mamba Modulation: On the Length Generalization of Mamba Models (NeurIPS 2025 · similarity 0.85)
- Understanding Input Selectivity in Mamba: Impact on Approximation Power, Memorization, and Associative Recall Capacity (ICML 2025 · similarity 0.85)
- Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality (ICML 2024 · similarity 0.84)
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.
