Abstract
Deep learning models have shown their strengths in various applicationdomains, however, they often struggle to meet safety requirements for theiroutputs. In this paper, we introduce PiShield, the first package ever allowingfor the integration of the requirements into the neural networks' topology.PiShield guarantees compliance with these requirements, regardless of input.Additionally, it allows for integrating requirements both at inference and/ortraining time, depending on the practitioners' needs. Given the widespreadapplication of deep learning, there is a growing need for frameworks allowingfor the integration of the requirements across various domains. Here, weexplore three application scenarios: functional genomics, autonomous driving,and tabular data generation.