Quick-Tune: Quickly Learning Which Pretrained Model to Finetune and How

Learn about a novel methodology for efficiently selecting and fine-tuning pretrained models in this 40-minute AutoML seminar presentation. Explore how to leverage knowledge transfer across multiple pretrained models and hyperparameter configurations using a comprehensive meta-dataset of over 20,000 evaluations spanning 24 image classification models and 87 datasets. Discover the implementation of a gray-box performance predictor for rapid hyperparameter optimization on new datasets, enabling faster and more accurate model selection. Gain practical insights from empirical demonstrations showing how to quickly identify and optimize the most suitable pretrained model for specific use cases. Access the complete research through the provided arXiv paper and explore the implementation details via the GitHub repository.