Researchers have introduced a novel approach called prior-fitted diffusion transformers for optimizing unknown functions in a black-box manner. This method utilizes diffusion models to perform Thompson sampling in-context after pre-training with the prior-fitted network paradigm. The work also develops practical non-Gaussian priors for functions with characteristics not achievable under Gaussianity, demonstrating improved performance over traditional Gaussian process pipelines on global optimization benchmarks. AI
IMPACT Introduces a novel method for black-box optimization that could enhance AI model training and hyperparameter tuning.
RANK_REASON The cluster contains a research paper detailing a new method for function optimization. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Diffusion Transformers
- Gaussian process
- prior-fitted diffusion transformers
- Thompson sampling
- transformers
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
