Research map / Architectures and efficiency / RNNs and normalization
Tensorized Projection for High-Dimensional Binary Embedding
Weixiang Hong, Jingjing Meng, Junsong Yuan
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 more TT cores and higher rank at similar parameter budget (6 cores TT-8 vs 2 cores TT-1) (ablation) | Tensorized Projection · DNN-4096, 1024-bit ANN search | mAP: 4096 vs 4224 parameters, TT-1 mAP almost 50% lower | Improved |
| increase number of TT cores (ablation) | Tensorized Projection · DNN-4096, 1024-bit ANN search | mAP | Improved |
| increase number of TT cores (ablation) | Tensorized Projection · DNN-4096, 1024-bit encoding | encoding time | Hurt |
| use well-proportioned instead of ill-proportioned input/output tensor shapes (ablation) | Tensorized Projection · DNN-4096, 1024-bit ANN search | mAP | Improved |
| vary penalty weight beta from 0.1 to 100 (ablation) | Tensorized Projection · DNN-4096 | mAP | No clear effect |
| use TP instead of SP (sparse projection) at 4096 bits | binary embedding · CIFAR-10 classification | accuracy: 78.6% -> 81.2% | Improved |
| use TP instead of FBE at 1024 bits | binary embedding · CIFAR-10 classification | accuracy: 79.7% -> 79.3% | Hurt |
| use TP instead of SP at 4096 bits | binary embedding · Holidays+1M retrieval | encoding time (ms): 3.12 -> 1.17 | Improved |
| increase code length from 8192 to 16384 bits | Tensorized Projection · Holidays+1M retrieval | mAP | No clear effect |
| use TP instead of FBE at 8192 bits | binary embedding · Holidays+1M retrieval | mAP: 49.3% -> 49.3% | No clear effect |
Closest papers by meaning
- Fried Binary Embedding for High-Dimensional Visual Features (CVPR 2017 · similarity 0.88)
- Coarse-To-Fine Tensor Trains for Compact Visual Representations (ICML 2024 · similarity 0.86)
- Faster Binary Embeddings for Preserving Euclidean Distances (ICLR 2021 · similarity 0.85)
- Attention-based Iterative Decomposition for Tensor Product Representation (ICLR 2024 · similarity 0.85)
- Cherry-Picking Gradients: Learning Low-Rank Embeddings of Visual Data via Differentiable Cross-Approximation (ICCV 2021 · similarity 0.84)
- Distribution-Aware Tensor Decomposition for Compression of Convolutional Neural Networks (NeurIPS 2025 · 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.
