Privacy Preserving Computation

Privacy-preserving computation (PPC) aims to enable collaborative data analysis and machine learning without revealing sensitive information. Current research focuses on improving the efficiency of techniques like secure multi-party computation (MPC), differential privacy (DP), and homomorphic encryption (HE), particularly for deep neural networks and large language models, often employing novel algorithms and architectures like PolyKervNets and PrivateLoRA to mitigate computational and communication overheads. The development of robust and efficient PPC methods is crucial for unlocking the potential of data-driven applications while safeguarding individual privacy and fostering trust in data sharing.