// ◉ Multimodal & Vision
SigLIP (Sigmoid Loss)
Replaces softmax contrastive loss with pairwise sigmoid loss.
SigLIP 2023 Intermediate core
IN PLAIN WORDS
Each image–text pair is judged independently without batch-wide normalization, working at small batches and scaling easily. A common vision encoder in open VLMs.
Used in: PaliGemma, Gemma 3 and other VLMs
01Original paper
Sigmoid Loss for Language Image Pre-Training
Xiaohua Zhai, Basil Mustafa, Alexander Kolesnikov, et al. · 2023
arXiv ↗ ✓ Verified · arXiv API
02Where it sits
Learn the left column first; the right column builds on this one.
SigLIP (Sigmoid Loss)
Then you can learn
- Nothing in the atlas builds on it yet
Related:Visual Instruction Tuning (LLaVA)
Left: prerequisites · right: builds on this · dashed: related. Full map →
03Learning resources
Official docs, explainers, courses and reference code. All links checked.
04Where the course teaches it
For AI / data
- Verification: verified, checked 2026-10-21.
- JSON: /api/learn/algorithms.json (id:
siglip)