Publication:
Saffe: multimodal model composition with semantic‑alignment fusion of frozen encoders

Loading...
Thumbnail Image
Date
2025-07-07
relationships.isAuthorOfPublication
relationships.isSecondaryAuthorOf
relationships.isDirectorOf
Authors
Kulasekara, Maithri ; Inglés‑Romero, Juan F. ; Imbernón, Baldomero ; Abellán, José L.
item.page.secondaryauthor
item.page.director
Publisher
Springer
publication.page.editor
DOI
https://doi.org/10.1007/s11227-025-07473-7
item.page.type
info:eu-repo/semantics/article
Description
© 2025, The Author(s). This manuscript version is made available under the CC-BY 4.0 license http://creativecommons.org/licenses/by/4.0/. This document is the Published version of a Published Work that appeared in final form in Journal of Supercomputing. To access the final edited and published work see https://doi.org/10.1007/s11227-025-07473-7
Abstract
Transformer-based multimodal models often require expensive, full-model training on task-specific all-modality datasets to achieve high accuracy on targeted downstream tasks. To reduce this significant cost, we introduce SAFFE, a methodology for building accurate, task-specific multimodal models with minimal training, using only standard GPU hardware. SAFFE leverages off-the-shelf, pre-trained, frozen unimodal encoders for each input modality (e.g., text, image, or audio) and connects them through a lightweight, trainable component called the FusionAlign Module (FAM). FAM is a bottleneck mid-fusion neural network, trained on the target data set to align the outputs of the independently pre-trained unimodal encoders. This approach eliminates the need for end-to-end training while maintaining strong accuracy for the downstream task. As a proof of concept, we validate SAFFE on image retrieval and language understanding tasks. SAFFE-derived models outperform state-of-the-art multimodal systems on datasets such as CIFAR-10, ImageNet-100, and COCO, achieving competitive results with significantly fewer trainable parameters and training time.
Citation
The Journal of Supercomputing (2025) 81:1114
item.page.embargo
Collections