Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
ExPT: Synthetic Pretraining for Few-Shot Experimental Design
Tung Nguyen, Sudhanshu Agrawal, Aditya Grover
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 |
|---|---|---|---|
| replace inverse modeling with forward modeling (TNP-ED + gradient ascent) (ablation) | ExPT · Design-Bench poorest, D'Kitty | mean score: 0.871 -> 0.405 | Hurt |
| replace inverse modeling with forward modeling (TNP-ED + gradient ascent) (ablation) | ExPT · Design-Bench random, Ant | mean score: 0.639 -> 0.451 | Hurt |
| make GP length scale too small (U[0.1, 1.0]) (ablation) | ExPT · Design-Bench random, Ant | mean score: 0.639 -> 0.518 | Hurt |
| make GP length scale too large (U[100, 200]) (ablation) | ExPT · Design-Bench random, Ant | mean score: 0.639 -> 0.528 | Hurt |
| make GP function scale too large (U[100, 200]) (ablation) | ExPT · Design-Bench random, Ant | mean score: 0.639 -> 0.534 | Hurt |
| make GP function scale too small (U[0.01, 0.1]) (ablation) | ExPT · Design-Bench D'Kitty mean score | mean score: 0.865 -> 0.896 (random); 0.871 -> 0.755 (poorest) | Mixed |
| pretrain on GP-Cosine instead of GP-RBF data (ablation) | ExPT · Design-Bench random, mean over 4 tasks | mean score: 0.614 -> 0.525 | Hurt |
| pretrain on trained-MLP generated functions instead of GP-RBF (ablation) | ExPT · Design-Bench random, mean over 4 tasks | mean score: 0.614 -> 0.625 | No clear effect |
| replace VAE decoder with diffusion decoder (ablation) | ExPT · Design-Bench random, mean over 4 tasks | mean score: 0.614 -> 0.575 | Hurt |
| replace VAE decoder with diffusion decoder (ablation) | ExPT · Design-Bench poorest, mean over 4 tasks | mean score: 0.639 -> 0.514 | Hurt |
| use sequential sampling with feedback instead of simultaneous sampling (ablation) | ExPT · Design-Bench poorest, Ant | mean score: 0.646 -> 0.767 | Improved |
| sort context and target points during pretraining (ablation) | ExPT · Design-Bench random, D'Kitty | mean score: 0.865 -> 0.794 | Hurt |
Closest papers by meaning
- Exploring Diffusion Transformer Designs via Grafting (NeurIPS 2025 · similarity 0.85)
- From Shortcut to Induction Head: How Data Diversity Shapes Algorithm Selection in Transformers (NeurIPS 2025 · similarity 0.84)
- Multiple Physics Pretraining for Spatiotemporal Surrogate Models (NeurIPS 2024 · similarity 0.84)
- Can Looped Transformers Learn to Implement Multi-step Gradient Descent for In-context Learning? (ICML 2024 · similarity 0.83)
- Born a Transformer -- Always a Transformer? On the Effect of Pretraining on Architectural Abilities (NeurIPS 2025 · similarity 0.83)
- Task-Specific Preconditioner for Cross-Domain Few-Shot Learning (AAAI 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.
