Privacy-preserving Contact Context Estimation

隐私保护接触上下文估计

• 批准号: 492351968
• 负责人: Professor Dr. Esfandiar Mohammadi
• 金额: --
• 依托单位: Klinik für Infektiologie und Mikrobiologie
• 依托单位国家: 德国
• 项目类别: Research Grants
• 财政年份: 2021
• 资助国家: 德国
• 起止时间: 2020-12-31 至 2022-12-31
• 项目状态: 已结题
• 来源: https://gepris.dfg.de/gepris/projekt/492351968?language=en
• 关键词: Privacy; preserving; Contact; Context; Estimation

Contact tracing apps provide data about whether a critical COVID-19 contact occurred. This contact tracing is solely based on Bluetoot signals. We plan to extend this mechanism with acustic indicators that incorporate the surrounding, i.e., the context of a critical contact. The recorded data would not leave the device yet privacy-preserving continuous learning of the acustic scene classifiers shall be done via differentially private federated learning. Complementarily, we plan to collect additional acustic data in a clinical study about coughing symtoms and change in speech by wearing a mask. As a result, we would validate publicly available audio data and extend the publicly available datasets for future research.

接触者追踪应用程序提供有关是否发生关键COVID-19接触的数据。这种联系人跟踪仅基于蓝牙信号。我们计划扩展这一机制与声学指标，包括周围，即，关键联系人的背景。记录的数据不会离开设备，但声学场景分类器的隐私保护连续学习应通过差分私有联合学习来完成。作为补充，我们计划在一项关于戴口罩引起咳嗽症状和言语变化的临床研究中收集更多的声学数据。因此，我们将验证公开可用的音频数据，并扩展公开可用的数据集以供未来研究。