pWCET-AI package pWCET-AI is an integrated solution for Probabilistic Timing Analysis of AI-based applications, fully integrated with SAFEXPLAIN middleware layer. It comprises a set of R scripts (RestK and TailID) that deploy cutting-edge techniques to cope with the limitations of most popular EVT-based methods when applied to AI-based SW functions. The latter are, in fact, much harder to analyse due to the presence of mixture distributions. See: - Sergi Vilardell, Isabel Serra, Enrico Mezzetti, Jaume Abella, Francisco J. Cazorla, Joan del Castillo: Using Markov's Inequality with Power-Of-k Function for Probabilistic WCET Estimation. ECRTS 2022: 20:1-20:24 - Blau Manau, Sergi Vilardell, Isabel Serra, Enrico Mezzetti, Jaume Abella, Francisco J. Cazorla: Detecting Low-Density Mixtures in High-Quantile Tails for pWCET Estimation. ECRTS 2025: 20:1-20:25* RestK https://cran.r-project.org/web/packages/RESTK/index.html TailID https://cran.r-project.org/web/packages/TailID/index.html SAFEXPLAIN Middleware integration package The integration of probabilistic timing methods is built on the capabilities of Orin-PMULib. The integration uses two additional middleware components: - PMULogger (smw_util.zip): Gathers PMU information collected through the PMULib and makes it available as a topic for the PMUVisualizer. - PMUVisualizer (smw_pmu_logger): Runs on a remote node in the same VPN as the main system and reads the information from the PMULogger. Once enough samples are collected, it runs the Restk.R script to obtain the pWCET distribution and plots the Cumulative Distribution Function for the probabilistic timing behavior against a given exceedance threshold.