AI-based anomaly detection on RISC-V

AI-based anomaly detection on RISC-V. This thesis aims at creating a powerful profiling infrastructure based on the latest specifications of the RISC-V HPM on a selected RISC-V core in order to provide PMCs in the architecture. The presence of PMCs provide aggregate information (e.g., number of memory transactions) that are useful to profile applications at run-time introducing a negligible overhead. The new profiling infrastructure provides to analyze softer and identify malicious modifications. However, the complexity of modern software still represents a challenge. Traditional tools for software verification and validation present limitations that may impact the quality of software. This thesis focuses on applying AI to support software developers in anomaly detection activities. The huge amount of data that can be collected resorting to PMCs is an important playground to train specific AI-based anomaly detection models able to identify both software bugs and security threats.