AINet: Anchor Instances Learning for Regional Heterogeneity in Whole Slide Image
Tingting Zheng, Hongxun Yao, Kui Jiang, Sicheng Zhao, Yi Xiao, Harbin Institute of Technology, Tsinghua University, Zhengzhou University · 2026
Read AINet: Anchor Instances Learning for Regional Heterogeneity in Whole Slide Image by Tingting Zheng, Hongxun Yao, Kui Jiang, Sicheng Zhao, Yi Xiao, Harbin Institute of Technology, Tsinghua University, Zhengzhou University (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-20187.txt.
Description
Multi-instance learning performs well on whole slide images, yet tumors are sparse and varied in shape, so regions differ markedly and pooling strong representations is hard. A 2026 arXiv paper by Tingting Zheng and four co-authors introduces anchor instances: a compact subset of instances that stand for their regions locally and stand out at the bag level globally.
Anchors serve as meaning guides for cross-region links. They correct weak patterns while preserving local diversity. A dual-level anchor mining module picks anchors from massive instance sets. In each region, the best anchor is picked by comparing its likeness to local and global embeddings. Further anchor-guided steps ensure the selected anchors are diverse and complete, giving whole slide image models region-aware references for pooling.
Catalog record
| CREATOR | Tingting Zheng, Hongxun Yao, Kui Jiang, Sicheng Zhao, Yi Xiao, Harbin Institute of Technology, Tsinghua University, Zhengzhou University |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20187