Research map / Architectures and efficiency / RNNs and normalization
Attentive Tensor Product Learning
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 |
|---|---|---|---|
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | METEOR: 0.257 -> 0.258 | Helped |
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | BLEU-1: 0.728 -> 0.733 | Helped |
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | BLEU-2: 0.566 -> 0.572 | Helped |
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | BLEU-3: 0.433 -> 0.437 | Helped |
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | BLEU-4: 0.330 -> 0.335 | Helped |
| replace the pure LSTM decoder with an attention-based TPR decoder that learns filler and role sequences | Image captioning decoder · MS COCO | CIDEr: 1.012 -> 1.013 | Helped |
| use unsupervisedly learned unbinding vectors as additional grammatical features | POS tagger · Penn TreeBank WSJ 22 | accuracy: 0.972 -> 0.973 | Helped |
| use unsupervisedly learned unbinding vectors as additional grammatical features | POS tagger · Penn TreeBank WSJ 23 | accuracy: 0.973 -> 0.974 | Helped |
| provide ground-truth tree-structure encodings instead of predicting them (ablation) | ATPL constituency parser · Penn TreeBank WSJ 22 | F-1 measure: 0.900 -> 0.963 | Helped |
| provide ground-truth tree-structure encodings instead of predicting them (ablation) | ATPL constituency parser · Penn TreeBank WSJ 23 | F-1 measure: 0.908 -> 0.965 | Helped |
| predict tree-structure encodings rather than use ground-truth encodings (ablation) | ATPL constituency parser · Penn TreeBank WSJ 22 | F-1 measure: 0.963 -> 0.900 | Hurt |
| predict tree-structure encodings rather than use ground-truth encodings (ablation) | ATPL constituency parser · Penn TreeBank WSJ 23 | F-1 measure: 0.965 -> 0.908 | Hurt |
Closest papers by meaning
- Attention-based Iterative Decomposition for Tensor Product Representation (ICLR 2024 · similarity 0.86)
- Tensor Product Attention Is All You Need (NeurIPS 2025 · similarity 0.85)
- Discrete Dictionary-based Decomposition Layer for Structured Representation Learning (NeurIPS 2024 · similarity 0.83)
- Fully Distributed, Flexible Compositional Visual Representations via Soft Tensor Products (NeurIPS 2024 · similarity 0.83)
- TNCSE: Tensor Norm Constraints for Unsupervised Contrastive Learning of Sentence Embeddings (AAAI 2025 · similarity 0.83)
- Training Neural Networks as Recognizers of Formal Languages (ICLR 2025 · similarity 0.83)
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.