It uses foundation model embeddings to handle data across various domains without forgetting previously learned information. This technology is particularly adept at categorizing images in a zero-shot learning context, where it assigns clusters to unseen data and applies experience replay to reinforce memory and prevent forgetting.
TADILER: Experience Replay in Zero-shot Clustering: Foundation Model Embeddings for Strategy-agnostic Domain Incremental Learning
Artificial Intelligence BSC Group: Computer Sciences SoftwareTADILER (Task-Agnostic Domain-Incremental Learning through Experience Replay), developed by Lenovo and BSC, is a methodology to derive AI models that employs zero-shot clustering and experience replay for domain-incremental learning.
Contacto:
License:
Apache License (Version 2.0)
Primary tabs
