Research map / Probabilistic and scientific modeling / Molecules and proteins
Beam Tree Recursive Cells
Jishnu Ray Chowdhury, Cornelia Caragea
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 |
|---|---|---|---|
| add beam search with plain top-k to GumbelTreeGRC | GumbelTreeGRC · ListOps length generalization (900-1000) | accuracy: 37.9% to 86.7% | Improved |
| add OneSoft top-k to BT-GRC (ablation) | BT-GRC · ListOps length generalization (900-1000) | accuracy: 86.7 -> 97.2 | Improved |
| add OneSoft top-k to BT-GRC (ablation) | BT-GRC · ListOps argument generalization (15 arguments) | accuracy: 63.7 -> 67.9 | Improved |
| add OneSoft top-k to BT-GRC (ablation) | BT-GRC · ListOps argument generalization (10 arguments) | accuracy: 77.1 -> 76.05 | Hurt |
| add OneSoft top-k to BT-GRC (ablation) | BT-GRC · ListOps-DG length generalization (900-1000) | accuracy: 99.9 -> 97.1 | Hurt |
| add OneSoft top-k to BT-GRC (ablation) | BT-GRC · Logical inference systematicity split C | accuracy: 94.17_{14} -> 78.04_{43} | Hurt |
| replace GRC with LSTM cell (ablation) | BT-Cell · ListOps length generalization (900-1000) | accuracy: 86.7 -> 78.8 | Hurt |
| replace OneSoft with SOFT Top-k (ablation) | BT-GRC + OneSoft · ListOps length generalization (900-1000) | accuracy: 97.2 -> 29.4 | Hurt |
| replace beam search with parallel Monte Carlo GumbelTree runs (ablation) | BT-GRC · ListOps length generalization (900-1000) | accuracy: 86.7 -> 25.1 | Hurt |
| reduce beam size from 5 to 2 (ablation) | BT-GRC · ListOps length generalization (900-1000) | accuracy: 86.7 -> 50.2 | Hurt |
| use OneSoft top-k with beam size 2 instead of plain top-k (ablation) | BT-GRC · ListOps length generalization (900-1000) | accuracy: 50.2 -> 91% | Improved |
| add beam search with plain top-k rather than greedy Gumbel-tree selection | GumbelTreeGRC · ListOps near-IID (length ≤1000) | accuracy: 74.89 -> 99.39 | Improved |
Closest papers by meaning
- Efficient Beam Tree Recursion (NeurIPS 2023 · similarity 0.92)
- Provable Test-Time Scaling for Beam Search in LLM Reasoning (NeurIPS 2026 · similarity 0.82)
- Boosting Inference with Guided Reasoning: Stochastic Exploration for Recursive Models (NeurIPS 2026 · similarity 0.82)
- Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA (ICLR 2025 · similarity 0.82)
- Augmenting Transformers with Recursively Composed Multi-grained Representations (ICLR 2024 · similarity 0.81)
- Entropy-Gated Latent Recursion (NeurIPS 2026 · similarity 0.81)
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.
