CoLLIE: Continual Learning of Language Grounding from Language-Image Embeddings
This addresses the challenge of adapting language-vision models to new language uses without retraining, which is incremental but useful for applications requiring flexible AI systems.
The paper tackles the problem of continual learning for language grounding in vision by introducing CoLLIE, which adjusts language embeddings from a pre-trained multimodal model to accommodate new language use with minimal interference, achieving efficient learning and generalization from few examples while preserving zero-shot performance.
This paper presents CoLLIE: a simple, yet effective model for continual learning of how language is grounded in vision. Given a pre-trained multimodal embedding model, where language and images are projected in the same semantic space (in this case CLIP by OpenAI), CoLLIE learns a transformation function that adjusts the language embeddings when needed to accommodate new language use. This is done by predicting the difference vector that needs to be applied, as well as a scaling factor for this vector, so that the adjustment is only applied when needed. Unlike traditional few-shot learning, the model does not just learn new classes and labels, but can also generalize to similar language use and leverage semantic compositionality. We verify the model's performance on two different tasks of identifying the targets of referring expressions, where it has to learn new language use. The results show that the model can efficiently learn and generalize from only a few examples, with little interference with the model's original zero-shot performance.