Research map / Learning theory and algorithms / Kernels and regression theory
Towards Regression-Free Neural Networks for Diverse Compute Platforms
Rahul Duggal, Hao Zhou, Shuo Yang, Jun Fang, Yuanjun Xiong, Wei Xia
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 |
|---|---|---|---|
| jointly search reference and target with REG-NAS instead of independent EffNet training | EfficientNet-B1 · ImageNet val (MobileNet-V3 space) | NFR (%): 4.25 -> 2.16 | Improved |
| REG-NAS instead of focal distillation (EffNet+FD) | EffNet+FD · ImageNet val (MobileNet-V3 space) | NFR (%): 3.25 -> 2.16 | Improved |
| REG-NAS instead of focal distillation (RN101+FD) | RN101+FD · ImageNet val (ResNet space) | NFR (%): 3.06 -> 1.57 | Improved |
| add architecture constraint (CAS) to reward R2 (ablation) | MB-R2-300 · ImageNet val (MobileNet-V3 space) | NFR (%): 2.25 -> 2.16 | Improved |
| add architecture constraint (CAS) to reward R2 (ablation) | RN-R2-3000 · ImageNet val (ResNet space) | NFR (%): 1.76 -> 1.57 | Improved |
| use reward R2 (accuracy and NFR) instead of R0 (accuracy only) (ablation) | MB-R0-300 · ImageNet val (MobileNet-V3 space) | NFR (%): 2.51 -> 2.25 | Improved |
| use reward R1 (NFR only) instead of R2 (ablation) | MB-R2-300 · ImageNet val (MobileNet-V3 space) | Top-1 accuracy (%): 76.90 -> 76.20 | Hurt |
| fine-tune with CE (ablation) | MB-R0-300 · ImageNet val (MobileNet-V3 space) | NFR (%): 2.51 -> 2.53 | No clear effect |
| fine-tune with CE (ablation) | RN-R0-3000 · ImageNet val (ResNet space) | NFR (%): 2.04 -> 2.47 | Hurt |
| fine-tune with FD loss (ablation) | RN-(R2+CAS)-3000 · ImageNet val (ResNet space) | NFR (%): 1.57 -> 0.96 | Improved |
| fine-tune with FD loss (ablation) | MB-(R2+CAS)-300 · ImageNet val (MobileNet-V3 space) | NFR (%): 2.16 -> 2.12 | Improved |
| search small to large instead of large to small (ablation) | REG-NAS · ImageNet val (MobileNet-V3 space) | NFR | Improved |
Closest papers by meaning
- Neural Networks Perform Sufficient Dimension Reduction (AAAI 2025 · similarity 0.81)
- Variation Spaces for Multi-Output Neural Networks: Insights on Multi-Task Learning and Network Compression (NeurIPS 2024 · similarity 0.81)
- Joker: Joint Optimization Framework for Lightweight Kernel Machines (ICML 2025 · similarity 0.81)
- Scalable Neural Network Kernels (ICLR 2024 · similarity 0.80)
- TMPNN: High-Order Polynomial Regression Based on Taylor Map Factorization (AAAI 2024 · similarity 0.79)
- FREE: Faster and Better Data-Free Meta-Learning (CVPR 2024 · similarity 0.79)
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.
