Approximate nearest neighbors search (ANNS) is widely used to find similar items in cloud-hosted vector databases. But as a new arXiv preprint notes, the operation has a privacy problem: the query vector sent to the server can reveal sensitive details about the user's intent. The paper, titled "Amora: Communication-Efficient Private Approximate Nearest Neighbors Search," proposes a private ANNS scheme called Amora to address this exposure.

According to the abstract, Amora sits within the broader private ANNS (PANNS) research area. The title emphasizes communication efficiency, suggesting that reducing the data exchanged between client and server is a core design goal. This matters because private search methods often add overhead through encryption or multiple rounds of interaction, which can be costly in practice.

The abstract provides only a high-level description; it does not detail Amora's underlying mechanism or present evaluation results. As a preprint, the work has not necessarily undergone peer review. Readers interested in how Amora achieves its efficiency will need to consult the full paper.