Research map / Architectures and efficiency / RNNs and normalization
Learning to Compose Task-Specific Tree Structures
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 |
|---|---|---|---|
| add LSTM-based leaf transformation (ablation) | Gumbel Tree-LSTM (100D) · SNLI | accuracy: 81.8 -> 82.6 | Helped |
| add LSTM-based leaf transformation (ablation) | Gumbel Tree-LSTM (300D) · SNLI | accuracy: 84.4 -> 85.6 | Helped |
| learn task-specific trees from plain text rather than using supplied parse-tree structures | Gumbel Tree-LSTM · SNLI | accuracy: Gumbel Tree-LSTM 300D: 85.6; Constituency-based and other RvNN baselines are lower | Helped |
| learn task-specific trees from plain text rather than using supplied parse-tree structures | Gumbel Tree-LSTM · SST-2 | accuracy: Gumbel Tree-LSTM: 90.7; Constituency Tree-LSTM: 88.0 | Helped |
| learn task-specific trees from plain text rather than using supplied parse-tree structures | Gumbel Tree-LSTM · SST-5 | accuracy: Gumbel Tree-LSTM: 53.7; Constituency Tree-LSTM: 51.0 | Helped |
| use Gumbel Tree-LSTM instead of Shortcut-Stacked BiLSTM | SNLI · SNLI | accuracy: 86.1 vs 86.0 | No clear effect |
| use Gumbel Tree-LSTM instead of BCN+Char+CoVe | Gumbel Tree-LSTM · SST-5 | accuracy: 53.7 vs 53.7 | No clear effect |
| use Gumbel Tree-LSTM instead of byte-mLSTM | Gumbel Tree-LSTM · SST-2 | accuracy: 91.8 vs 90.7 | Hurt |
| use Gumbel Tree-LSTM instead of CYK Tree-LSTM | Gumbel Tree-LSTM (100D) · SNLI | training time: 240 -> 0.6 | Helped |
Closest papers by meaning
- Dynamic Compositionality in Recursive Neural Networks with Structure-Aware Tag Representations (AAAI 2019 · similarity 0.88)
- Tree-structured decoding with doubly-recurrent neural networks (ICLR 2017 · similarity 0.87)
- Modeling Hierarchical Structures with Continuous Recursive Neural Networks (ICML 2021 · similarity 0.87)
- Tree-Structured Attention with Hierarchical Accumulation (ICLR 2020 · similarity 0.86)
- Learning to Embed Sentences Using Attentive Recursive Trees (AAAI 2019 · similarity 0.86)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · 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.