<ul><li>Bayesian optimization (BO) aims to optimize expensive black-box functions in science and engineering.</li><li>A new approach is proposed that eliminates the need for re-training the surrogate model and optimizing the acquisition function at each iteration.</li><li>This method uses a pre-trained deep generative model to directly sample from the posterior over the optimum point, achieving efficiency gains of over 35x compared to traditional methods.</li><li>The approach allows for efficient parallel and distributed BO, especially beneficial for high-throughput optimization tasks.</li></ul>

Simplifying Bayesian Optimization Via In-Context Direct Optimum Sampling

Discover more