AutoSimulate: (Quickly) learning synthetic data generation
Simulation is increasingly being used for generating large labelled datasets in many machine learning problems. Recent methods have focused on adjusting simulator parameters with the goal of maximising accuracy on a validation task, usually relying on REINFORCE-like gradient estimators. However thes...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Springer
2020
|
| _version_ | 1797080050118426624 |
|---|---|
| author | Behl, HS Baydin, AG Gal, R Torr, PHS Vineet, V |
| author_facet | Behl, HS Baydin, AG Gal, R Torr, PHS Vineet, V |
| author_sort | Behl, HS |
| collection | OXFORD |
| description | Simulation is increasingly being used for generating large labelled datasets in many machine learning problems. Recent methods have focused on adjusting simulator parameters with the goal of maximising accuracy on a validation task, usually relying on REINFORCE-like gradient estimators. However these approaches are very expensive as they treat the entire data generation, model training, and validation pipeline as a black-box and require multiple costly objective evaluations at each iteration. We propose an efficient alternative for optimal synthetic data generation, based on a novel differentiable approximation of the objective. This allows us to optimize the simulator, which may be non-differentiable, requiring only one objective evaluation at each iteration with a little overhead. We demonstrate on a state-of-the-art photorealistic renderer that the proposed method finds the optimal data distribution faster (up to 50×), with significantly reduced training data generation and better accuracy on real-world test datasets than previous methods. |
| first_indexed | 2024-03-07T00:54:34Z |
| format | Conference item |
| id | oxford-uuid:87962429-2603-446b-8ca2-afdf055a0088 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T00:54:34Z |
| publishDate | 2020 |
| publisher | Springer |
| record_format | dspace |
| spelling | oxford-uuid:87962429-2603-446b-8ca2-afdf055a00882022-03-26T22:11:45ZAutoSimulate: (Quickly) learning synthetic data generationConference itemhttp://purl.org/coar/resource_type/c_5794uuid:87962429-2603-446b-8ca2-afdf055a0088EnglishSymplectic ElementsSpringer2020Behl, HSBaydin, AGGal, RTorr, PHSVineet, VSimulation is increasingly being used for generating large labelled datasets in many machine learning problems. Recent methods have focused on adjusting simulator parameters with the goal of maximising accuracy on a validation task, usually relying on REINFORCE-like gradient estimators. However these approaches are very expensive as they treat the entire data generation, model training, and validation pipeline as a black-box and require multiple costly objective evaluations at each iteration. We propose an efficient alternative for optimal synthetic data generation, based on a novel differentiable approximation of the objective. This allows us to optimize the simulator, which may be non-differentiable, requiring only one objective evaluation at each iteration with a little overhead. We demonstrate on a state-of-the-art photorealistic renderer that the proposed method finds the optimal data distribution faster (up to 50×), with significantly reduced training data generation and better accuracy on real-world test datasets than previous methods. |
| spellingShingle | Behl, HS Baydin, AG Gal, R Torr, PHS Vineet, V AutoSimulate: (Quickly) learning synthetic data generation |
| title | AutoSimulate: (Quickly) learning synthetic data generation |
| title_full | AutoSimulate: (Quickly) learning synthetic data generation |
| title_fullStr | AutoSimulate: (Quickly) learning synthetic data generation |
| title_full_unstemmed | AutoSimulate: (Quickly) learning synthetic data generation |
| title_short | AutoSimulate: (Quickly) learning synthetic data generation |
| title_sort | autosimulate quickly learning synthetic data generation |
| work_keys_str_mv | AT behlhs autosimulatequicklylearningsyntheticdatageneration AT baydinag autosimulatequicklylearningsyntheticdatageneration AT galr autosimulatequicklylearningsyntheticdatageneration AT torrphs autosimulatequicklylearningsyntheticdatageneration AT vineetv autosimulatequicklylearningsyntheticdatageneration |