Key points are not available for this paper at this time.
With the emergence of Transformers and Vision-Language Models (VLMs) such as CLIP, fine-tuning large pre-trained models has recently become a prevalent strategy in Continual Learning. This has led to the development of numerous prompting strategies to adapt transformer-based models without incurring catastrophic forgetting. However, these strategies often compromise the original zero-shot capabilities of the pre-trained CLIP model and struggle to adapt to domains that significantly deviate from the pre-training data. In this work, we propose Continual Generative training for Incremental prompt-Learning, a simple and novel approach to mitigate forgetting while adapting CLIP. Briefly, we employ Variational Autoencoders (VAEs) to learn class-conditioned distributions within the embedding space of the visual encoder. We then exploit these distributions to sample new synthetic visual embeddings and train the corresponding class-specific textual prompts during subsequent tasks. Through extensive experiments on different domains, we show that such a generative replay approach can adapt to new tasks while improving zero-shot capabilities, evaluated using a novel metric tailored for CL scenarios. Notably, further analysis reveals that our approach can bridge the gap with joint prompt tuning. The codebase is available at https://github.com/aimagelab/mammoth.
Building similarity graph...
Analyzing shared references across papers
Loading...
Frascaroli et al. (Mon,) studied this question.
www.synapsesocial.com/papers/68e5f93bb6db64358758d72c — DOI: https://doi.org/10.48550/arxiv.2407.15793
Emanuele Frascaroli
Aniello Panariello
Pietro Buzzega
Building similarity graph...
Analyzing shared references across papers
Loading...