Research map / Architectures and efficiency / RNNs and normalization
Linking Image and Text With 2-Way Nets
Check 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 |
|---|---|---|---|
| Use the tied 2-way network with Euclidean loss and variance-preserving components | 2WayNet · XRMB | sum correlation: 110.18 vs 107.9 | Helped |
| Remove batch normalization (ablation) | 2WayNet · MNIST | correlation: 49.15 -> 13.14 | Hurt |
| Remove batch normalization (ablation) | 2WayNet · Flickr8k | recall@1 image search: 29.3 -> 21.1 | Hurt |
| Apply BN after ReLU rather than before ReLU (ablation) | 2WayNet · Flickr8k | correlation: 1313 -> 1758 | Helped |
| Add inverse-scale variance regularization Rγ (ablation) | 2WayNet · Flickr8k | correlation: 1598 -> 1758 | Helped |
| Use tied dropout instead of conventional dropout (ablation) | 2WayNet · Flickr8k | correlation: 1557 -> 1758 | Helped |
| Remove dropout (ablation) | 2WayNet · Flickr8k | correlation: 1758 -> 1091 | Hurt |
| Replace leaky ReLU with conventional ReLU (ablation) | 2WayNet · Flickr8k | correlation: about 33% (1192 total correlation) | Hurt |
| Remove the mid-network matching loss Lh (ablation) | 2WayNet · Flickr8k | correlation: 31% (1230) decrease | Hurt |
| Remove both reconstruction losses Lx and Ly (ablation) | 2WayNet · Flickr8k | correlation: increases by 56% (2752) | Helped |
| Remove both reconstruction losses Lx and Ly (ablation) | 2WayNet · Flickr8k image search | recall@1: 29.3 to 4.0; decrease of 87% | Hurt |
| Remove decorrelation regularization Rdecov (ablation) | 2WayNet · Flickr8k image search | recall@1: decrease by 14% | Hurt |
Closest papers by meaning
- Unsupervised Correlation Analysis (CVPR 2018 · similarity 0.90)
- Scalable and Effective Deep CCA via Soft Decorrelation (CVPR 2018 · similarity 0.86)
- Unpaired Canonical Correlation Analysis (NeurIPS 2026 · similarity 0.85)
- On Sparse Canonical Correlation Analysis (NeurIPS 2024 · similarity 0.84)
- MATCHA: Towards Matching Anything (CVPR 2025 · similarity 0.84)
- FC-GNN: Recovering Reliable and Accurate Correspondences from Interferences (CVPR 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.