A new arXiv preprint, DISTA-LP, focuses on linear-programming (LP) mechanisms for metric differential privacy (mDP). According to the abstract, these mechanisms are valuable because they enable fine-grained, task-aware control over utility, but they are computationally expensive and become privacy-fragile in distributed settings.
The announcement does not include the full description of the privacy risk, but the title and framing indicate the proposed response is a distributed LP design. The authors introduce DISTA-LP as a privacy-preserving distributed linear-programming method, presumably to keep the utility benefits of LP mechanisms while avoiding the vulnerabilities of centralised data handling.
Because the available text is only the abstract, the article cannot assess experimental results or compare DISTA-LP with alternative approaches. The source is a single announcement, so there are no independent findings to corroborate or contrast with the authors' claims.