ASIF: coupled data turns unimodal models to multimodal without training

Antonio Norelli, Marco Fumero, Valentino Maiorca, Luca Moschella, Emanuele Rodolà, Francesco Locatello

Keywords: Representation learning, Multimodal models, Analogy, Sparsity, Relative representations

TL;DR: How to build a CLIP-like model with two pretrained encoders and a limited amount of image-text pairs without tuning a neuron.

Abstract: Aligning the visual and language spaces requires to train deep neural networks from scratch on giant multimodal datasets; CLIP trains both an image and a text encoder, while LiT manages to train just the latter by taking advantage of a pretrained vision network. In this paper, we show that sparse relative representations are sufficient to align text and images without training any network. Our method relies on readily available single-domain encoders (trained with or without supervision) and a modest (in comparison) number of image-text pairs. ASIF redefines what constitutes a multimodal model by explicitly disentangling memory from processing: here the model is defined by the embedded pairs of all the entries in the multimodal dataset, in addition to the parameters of the two encoders. Experiments on standard zero-shot visual benchmarks demonstrate the typical transfer ability of image-text models. Overall, our method represents a simple yet surprisingly strong baseline for foundation multi-modal models, raising important questions on their data efficiency and on the role of retrieval in machine learning.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning