Research map / Architectures and efficiency / RNNs and normalization
Neural Parameter Allocation Search
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply SSN parameter sharing at a 6.9% parameter budget | EmbNet · Flickr30K | average Recall@K: 56.5 -> 67.9 | Helped |
| Apply SSN parameter sharing at a 34.7% parameter budget | ADAPT-T2I · Flickr30K | average Recall@K: 76.6 -> 81.2 | Helped |
| Apply SSN parameter sharing at a 7.8% parameter budget | SimNet · Flickr30K Entities | localization accuracy: 70.6 -> 71.9 | Helped |
| Use layer embeddings rather than an unshared reduced-size model to combine templates (ablation) | WRN-50-2 · ImageNet | Error@5: 10.08 -> 6.69 | Helped |
| Use Emb rather than WAvg to combine templates (ablation) | WRN-50-2 · ImageNet | Error@5: 7.38 -> 6.69 | Helped |
| Use learned masks on repeated parameter tiles instead of bilinear interpolation (ablation) | WRN-28-10 · CIFAR-100 | Error@1: 26.84 -> 25.71 | Helped |
| Use two parameter groups with mask upsampling instead of one group with mask upsampling (ablation) | WRN-28-10 · CIFAR-100 | Error@1: 25.71 -> 25.20 | Helped |
| Use learned parameter sharing rather than a hypernetwork at the same parameter budget | Wide ResNet 40-1 · CIFAR-10 | Error@1: 8.02 -> 7.22 | Helped |
| Increase parameter groups from 4 to 8 with 4M parameters (ablation) | SSN-WRN-28-10 · CIFAR-100 | Error@1: 21.31 $\pm$ 0.13 -> 22.01 $\pm$ 0.26 | Hurt |
| Vary template count from 4 to 10 (ablation) | SSN-EmbNet · Flickr30K | average Recall@K: 74.3 $\pm$ 0.25 -> 74.1 $\pm$ 0.16 | No clear effect |
| Use automatically learned parameter-group mappings rather than manual groups (ablation) | SSN · Image-sentence retrieval, phrase grounding, and image classification | task performance | No clear effect |
Closest papers by meaning
- Efficient Neural Architecture Search via Parameters Sharing (ICML 2018 · similarity 0.88)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.86)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.86)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.86)
- Principled Architecture-aware Scaling of Hyperparameters (ICLR 2024 · similarity 0.85)
- Encodings for Prediction-based Neural Architecture Search (ICML 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.