JAM aligns frozen vision and language models via joint autoencoders and multimodal Spread Loss, reliably inducing cross-modal alignment across layer depths, objectives, and model scales.
Learning transferable visual models from natural language supervision
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Escaping Plato's Cave: JAM for Aligning Independently Trained Vision and Language Models
JAM aligns frozen vision and language models via joint autoencoders and multimodal Spread Loss, reliably inducing cross-modal alignment across layer depths, objectives, and model scales.