KDD 2023 - Deep Pipeline Embeddings for AutoML
Sebastian Pineda Arango, University of Freiburg
How can one effectively search for Machine Learning and Deep Learning Pipelines? Typically, pipelines contain numerous conditional hyperparameters and correlated features. Moreover, they often result in large search spaces. We propose learning an embedding function that enables a more efficient search. This function is implemented using a neural network, which can be meta-learned or designed based on knowledge of the pipeline structure. We demonstrate that this approach outperforms the state-of-the-art method. Additionally, it can be easily adapted to new components added to the pipeline.
Смотрите видео KDD 2023 - Deep Pipeline Embeddings for AutoML онлайн без регистрации, длительностью часов минут секунд в хорошем качестве. Это видео добавил пользователь Association for Computing Machinery (ACM) 12 Июль 2023, не забудьте поделиться им ссылкой с друзьями и знакомыми, на нашем сайте его посмотрели 118 раз и оно понравилось 5 людям.