Research map / Architectures and efficiency / Neural architecture search
Expressive yet Efficient Feature Expansion with Adaptive Cross-Hadamard Products
Xuyang Zhang, Xi Zhang, Liang Chen, Hao Shi, Qingshan Guo
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 |
|---|---|---|---|
| use softsign instead of sigmoid in dynamic normalization (ablation) | ACH small model · CIFAR-100 | top-1 accuracy (%): 73.14 -> 73.57 | Improved |
| use softsign instead of algebraic sigmoid (ablation) | ACH small model · CIFAR-100 | top-1 accuracy (%): 72.80 -> 73.57 | Improved |
| replace IB9 and IB10 with ACH (ablation) | MobileNetV3-S · CIFAR-100 | top-1 accuracy (%): 70.01 -> 71.58 | Improved |
| replace only IB0 with ACH (ablation) | MobileNetV3-S · CIFAR-100 | top-1 accuracy (%): 70.01 -> 69.74 | Hurt |
| replace only IB1 with ACH (ablation) | MobileNetV3-S · CIFAR-100 | top-1 accuracy (%): 70.01 -> 69.74 | Hurt |
| replace last two layers with ACH (ablation) | MobileNetV3-S · CIFAR-100 | top-1 accuracy (%): 70.01 -> 71.58 | Improved |
| replace last two layers with ACH (ablation) | ShuffleNetV2-1.0 · CIFAR-100 | top-1 accuracy (%): 65.89 -> 71.68 | Improved |
| replace last two layers with ACH (ablation) | StarNet-S1 · CIFAR-100 | top-1 accuracy (%): 71.84 -> 72.07 | Improved |
| replace last two layers with ACH (ablation) | MobileNetV4-S · CIFAR-100 | top-1 accuracy (%): 73.15 -> 72.19 | Hurt |
| replace ACH modules in MobileViT-small (ablation) | MobileViT-small · ImageNet-1k | top-1 accuracy (%): 71.70 -> 72.42 | Improved |
| replace ACH modules in MobileViT-xx-small (ablation) | MobileViT-xx-small · ImageNet-1k | top-1 accuracy (%): 67.62 -> 67.42 | No clear effect |
| use parity-balanced indexing instead of native ACH | GPU ACH kernel · GPU latency | latency (ms): 12.40 -> 7.13 | Improved |
Closest papers by meaning
- Hadamard Representation: Scaffolding Performance Across Model-free RL (NeurIPS 2026 · similarity 0.84)
- Diagonalizing the Softmax: Hadamard Initialization for Tractable Cross-Entropy Dynamics (NeurIPS 2026 · similarity 0.82)
- HOT: Hadamard-based Optimized Training (CVPR 2025 · similarity 0.82)
- Implicit Modeling for Transferability Estimation of Vision Foundation Models (NeurIPS 2025 · similarity 0.81)
- An Adaptive Orthogonal Convolution Scheme for Efficient and Flexible CNN Architectures (ICML 2025 · similarity 0.81)
- LLMind: Bio-inspired Training-free Adaptive Visual Representations for Vision-Language Models (CVPR 2026 · similarity 0.80)
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.
