Formal Verification of Deep Neural Networks for Code Conversion applications

This ia a mobility Project funded in Austria as a cooperation between Nile University and Technical University of Vienna (TU WIEN) Deep Neual Networks are one of the most influential artificial intelligences technologies, and are used in many real-world scenarios. However, the magical nature of these networks has drawn a lot of criticism. It is disconcerting that we don't completely comprehend how this technology's decisionmaking works. A promising approach is to formally specify the requirements that a neural network must meet and employ formal verification to ensure the correctness of the decisions made by DNN. Incorporating formal verification would lead to an acceptable level of trust in the neural network decisions. The need for code conversion is increasing in different areas as a powerful technique for automatically migrating applications across different platforms. A smart accurate code conversion framework would ease the process of developing applications for different operating systems and help smooth migration to updated hardware technologies with better resources. The aim of this work is to provide a smart verified correct by construction code conversion platform. The developed platform is based on a deep learning system to enhance the conversion success rate by recognizing statements and predict the corresponding ones. Formal models and formal verification tools will be investigated to select the most suitable formal verification methodology to ensure increase the trust level in the DNN output and develop a correct by construction system