Research map / Probabilistic and scientific modeling / Bayesian optimization
Neural network gradient-based learning of black-box function interfaces
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 |
|---|---|---|---|
| Use Estimate and Replace instead of end-to-end baseline training | Text-Logic · Text-Logic test set (250 train examples) | accuracy: 0.533 -> 0.966 | Helped |
| Use Estimate and Replace instead of an end-to-end NALU | Image-Addition · MNIST test set (k=10) | mean absolute error: 1.42 -> 0.42 | Helped |
| Use Estimate and Replace instead of an end-to-end NALU | Image-Addition · MNIST test set (k=100) | mean absolute error: 7.88 -> 3.3 | Helped |
| Replace the estimator with the original lookup function at inference (ablation) | Image-Lookup EstiNet · MNIST test set (k=3) | accuracy: 0.1 -> 0.97 | Helped |
| Use hybrid training rather than offline training with a frozen estimator (ablation) | Text-Lookup-Logic EstiNet · inference | accuracy: 0.17 -> 0.98 | Helped |
| Use hybrid training rather than online training from a cold start (ablation) | Text-Lookup-Logic EstiNet · inference | accuracy: 0.69 -> 0.98 | Helped |
| Pretrain the estimator offline before joint training (ablation) | Text-Lookup-Logic EstiNet · train | accuracy: 0.76 -> 0.98 | Helped |
| Use the estimator in test mode instead of replacing it with the lookup function (ablation) | Image-Lookup EstiNet · MNIST test set (k=2) | accuracy: 0.97 -> 0.11 | Hurt |
| Regularize estimator confidence to preserve useful gradients (ablation) | Image-Lookup EstiNet · Image-Lookup (k=3) | argument-extractor gradient norm | Helped |
| Train with the full task dataset rather than a small fraction | Text-Logic EstiNet · Text-Logic test set | accuracy: 0.966 vs 0.533 | Mixed |
Closest papers by meaning
- Data-Efficient Learning with Neural Programs (NeurIPS 2024 · similarity 0.86)
- Network Inversion of Convolutional Neural Nets (Student Abstract) (AAAI 2025 · similarity 0.85)
- Transformers Implement Functional Gradient Descent to Learn Non-Linear Functions In Context (ICML 2024 · similarity 0.84)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.84)
- Natural Perturbations for Black-box Training of Neural Networks by Zeroth-Order Optimization (ICML 2025 · similarity 0.84)
- Towards Universal Offline Black-Box Optimization via Learning Language Model Embeddings (ICML 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.