Reproduce “Fusion Anything: A Generalized Multimodal Foundation Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Huizi Cui, Zongbo Han, Chenggong Ding, Naichuan Xiao, Jialong Yang, Jingdong Chen, Guangyu Wang, Qinghua Hu and 1 more
FAM uses synthetic multimodal tasks with varied cross-modal relationships to learn how modalities inform predictions. At inference, it combines pretrained modality representations with labeled examples from a new task, predicting without task-specific parameter updates. Across 18 datasets, it reports an ACC improvement of 6.92% over specialized fusion baselines. Results include gains and ties or lower rankings on some datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.