You can refer to the following methods, which are effective for adaptive sampling and can improve training efficiency in generative models.
Importance Sampling: You can adjust sampling based on data relevance, focusing on samples with higher error or significance.
The code snippet below shows how it is done:
Curriculum Learning: Starts with simpler samples, and then it gradually increases complexity as the model learns.
The code snippet below shows how it is done:
Uncertainty-Based Sampling: Prioritizes samples where the model shows high uncertainty, encouraging better learning on hard samples.
The code snippet below shows how it is done:
Hence, using these methods is effective for adaptive sampling and improving training efficiency in generative models.