Research map / Probabilistic and scientific modeling / Bayesian optimization
AutoSimulate: (Quickly) Learning Synthetic Data Generation
Harkirat Singh Behl, Atilim Güneş Baydin, Ran Gal, Philip H.S. Torr, Vibhav Vineet
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 |
|---|---|---|---|
| use the real validation set through the Newton-step approximation (vs No Validation) (ablation) | AutoSimulate · LM-O object detection | test mAP@0.50: 29.3 -> 36.1 | Improved |
| use linear approximation of the inner problem instead of exact quadratic (ablation) | AutoSimulate · LM-O object detection | test mAP@0.50: 36.1 -> 45.9 | Improved |
| use approximate quadratic instead of exact quadratic (ablation) | AutoSimulate · LM-O object detection | simulator training time (s): 23,193 -> 2,321 | Improved |
| use approximate quadratic instead of exact quadratic (ablation) | AutoSimulate · LM-O object detection | images generated: 8,950 -> 2,950 | Improved |
| use linear approximation instead of approximate quadratic (ablation) | AutoSimulate · LM-O object detection | simulator training time (s): 2,321 -> 30,477 | Hurt |
| freeze 98 of 112 layers of pretrained Yolo-spp (ablation) | AutoSimulate · LM-O object detection | test mAP@0.50: 45.9 -> 37.1 | Hurt |
| freeze 98 of 112 layers of pretrained Yolo-spp (ablation) | AutoSimulate · LM-O object detection | simulator training time (s): 30,477 -> 2,321 | Improved |
| freeze 104 of 112 layers (ablation) | AutoSimulate · LM-O object detection | test mAP@0.50: 45.9 -> 35.8 | Hurt |
| replace REINFORCE (LTS) with AutoSimulate | simulator optimizer · CLEVR segmentation | images generated: 3,750 -> 390 | Improved |
| replace Bayesian optimization with AutoSimulate | simulator optimizer · CLEVR segmentation | time: 43m -> 53m | Hurt |
| replace REINFORCE (LTS) with AutoSimulate (Linear) | simulator optimizer · LM-O object detection | test mAP@0.50: 37.2 -> 45.9 | Improved |
| replace REINFORCE (LTS) with AutoSimulate (Approx Quad) | simulator optimizer · LM-O object detection | simulator training time (s): 114,360 -> 2,321 | Improved |
Closest papers by meaning
- Improving the Scaling Laws of Synthetic Data with Deliberate Practice (ICML 2025 · similarity 0.86)
- Real-Fake: Effective Training Data Synthesis Through Distribution Matching (ICLR 2024 · similarity 0.85)
- Efficient Dataset Distillation for Pre-Trained Self-Supervised Models via Statistical Flow Matching (NeurIPS 2026 · similarity 0.84)
- Provably Improving Generalization of Few-shot models with Synthetic Data (ICML 2025 · similarity 0.84)
- On the Stability of Iterative Retraining of Generative Models on their own Data (ICLR 2024 · similarity 0.84)
- Increasing the Utility of Synthetic Images through Chamfer Guidance (NeurIPS 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.
