Description
AI systems that process rich input signals often transmit intermediate representations to the cloud, risking the exposure of sensitive user attributes such as identity or emotional state. Privacy-preserving signal processing methods that transform these representations to retain essential task-relevant information while suppressing private cues can help to improve user privacy. The approach is efficient, adaptable to different input types, and suitable for deployment in edge-to-cloud pipelines.

Task
Develop and evaluate a neural network-based method to suppress user-sensitive information from locally extracted representations while keeping a satisfying task performance.

Prerequisites

• Strong Python skills
• Solid knowledge about acoustics and signal processing
• Interest in artificial intelligence and deep neural networks