Embedding Inversion via Conditional Masked Diffusion: recover original text from embedding vectors using parallel denoising. Live demo + training pipeline + technical report.
-
Updated
Mar 7, 2026 - Python
Embedding Inversion via Conditional Masked Diffusion: recover original text from embedding vectors using parallel denoising. Live demo + training pipeline + technical report.
🛠 Reconstruct original text from text embeddings using conditional masked diffusion to reveal reversible embedding representations efficiently and accurately
Gradient-guided embedding inversion against BAAI/bge-m3 — proof that a stored embedding vector is not anonymized data.
State-of-the-art privacy for multi-vector VLM retrievers (ColPali/ColQwen2): a field-level PII linkage attack, the holographic leakage mechanism, an adaptive break of naive defenses, and Cataract — the first adaptively-evaluated, tunable, index-time real-privacy defense for this class.
Add a description, image, and links to the embedding-inversion topic page so that developers can more easily learn about it.
To associate your repository with the embedding-inversion topic, visit your repo's landing page and select "manage topics."