SARDINE

Artificial intelligence (AI) is increasingly permeating our everyday lives. Whether we ask ChatGPT for help with drafting documents at work or ask Apple's HealthGPT chatbot for advice on our health problems. As practical as the use of AI is, we must not forget that we often have to share our data with third parties to use it. In the above examples, this could mean sharing company secrets with ChatGPT or sensitive medical information with Apple. This raises not only the classic security problem of how to transfer my data to others in encrypted form, but also the new security challenge of how to have my data evaluated by others without the evaluator seeing it in plain text. Fortunately, there is a wide range of technologies available to solve these two problems. Our group's goal is to translate these security technologies from theory into practice. This involves implementing a variety of security mechanisms on actual hardware, whether it be resource-constrained IoT devices or powerful cloud servers, and assessing their performance. Currently, we are focused on privacy-preserving technologies such as homomorphic encryption, secure multi-party computation, and functional encryption. In simpler terms, these technologies enable calculations to be performed on data without sharing it in plain text with others.

• Homomorphic Encryption: Several libraries, such as OpenFHE and Zama, implement a wide range of homomorphic cryptosystems, including CKKS, BGV, and TFHE. These libraries allow operations on various encrypted data types, such as double, integer, and boolean. Potential student projects in this area could involve using existing libraries to implement additional homomorphic algorithms and evaluating their performance. 
• Concrete topics for thesis can be found at the chair's open thesis page.