Research map / Label-efficient and robust learning / Recognition and long-tail
Compositional Generalization via Neural-Symbolic Stack Machines
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 the symbolic stack machine and operational-equivalence category predictors | NeSS · SCAN Length | accuracy: around 20% -> 100.0 | Helped |
| Remove the latent category predictors (ablation) | NeSS · SCAN Length | accuracy: 100% in 2 runs; around 20% for other 3 runs | Hurt |
| Use a differentiable stack LSTM instead of the symbolic stack machine with sequence manipulation | SCAN Length · SCAN Length | accuracy: 100.0 vs 17.0 | Hurt |
| Use Synth greedy decoding instead of its grammar search | Synth · SCAN Length | accuracy: 0.0 vs 100.0 | Hurt |
| Use Synth greedy decoding instead of its grammar search | Synth · SCAN Around right | accuracy: 0.0 vs 100.0 | Hurt |
| Use Synth greedy decoding instead of its grammar search | Synth · SCAN Jump | accuracy: 3.5 vs 100.0 | Hurt |
| Use Synth greedy decoding instead of its grammar search | Synth · SCAN Simple | accuracy: 13.3 vs 100.0 | Hurt |
| Use a standard seq2seq model instead of NeSS | Compositional machine translation · Compositional machine translation | semantic equivalence accuracy: 12.5 vs 100.0 | Hurt |
| Use a standard seq2seq model instead of NeSS | Few-shot compositional instructions · Few-shot compositional instructions | accuracy: 2.5 vs 100.0 | Hurt |
| Use a standard seq2seq model instead of NeSS | While-Lang curriculum parsing · While-Lang Test-5000 (Curriculum) | accuracy: 0% vs 100% | Hurt |
| Use a standard seq2seq model instead of NeSS | Lambda-Lang curriculum parsing · Lambda-Lang Test-5000 (Curriculum) | accuracy: 0% vs 100% | Hurt |
Closest papers by meaning
- Neural-Symbolic Recursive Machine for Systematic Generalization (ICLR 2024 · similarity 0.90)
- When does compositional structure yield compositional generalization? A kernel theory. (ICLR 2025 · similarity 0.87)
- NeSyCoCo: A Neuro-Symbolic Concept Composer for Compositional Generalization (AAAI 2025 · similarity 0.86)
- ExeDec: Execution Decomposition for Compositional Generalization in Neural Program Synthesis (ICLR 2024 · similarity 0.86)
- Scalable Evaluation and Neural Models for Compositional Generalization (NeurIPS 2025 · similarity 0.86)
- CraftFactory: A Conditioned Control Policy Benchmark for Compositional Generalization (AAAI 2025 · similarity 0.86)
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.