TY  - RPRT
TI  - Bodhi VLM: Privacy-Alignment Modeling for Hierarchical Visual Representations in Vision Backbones and VLM Encoders via Bottom-Up and Top-Down Feature Search
AU  - Bo Ma
AU  - Wei Qi Yan
AU  - Jinsong Wu
PY  - 2026
UR  - https://arxiv.org/abs/2603.13728
ID  - 2603.13728
ER  -