DAQ-MPQ Assets: Checkpoints, Calibration Statistics, Results, and Figures

DAQ-MPQ external asset package This archive accompanies the DAQ-MPQ code repository and contains the external files required to reproduce the reported experiments. The package is organized into four top-level directories: checkpoints, stats, results, and figures. The checkpoints directory stores the baseline model weights used for evaluation. It includes the CIFAR-10 MobileNetV2 baseline checkpoint and the CIFAR-10 ResNet18/50 baseline checkpoints. The CIFAR-10 ResNet18 and ResNet50 checkpoints are community-provided pretrained baseline weights obtained from the GitHub repository `huyvnphan/PyTorch_CIFAR10`, and are used for the residual-backbone generalization experiments. ImageNet MobileNetV2 experiments use the official torchvision pretrained ImageNet weights and therefore do not require a separate custom MobileNetV2 checkpoint in this asset package. The stats directory stores offline calibration statistics consumed by DAQ, A-MPQ, and DAQ-MPQ related scripts. These files include percentile-based activation summaries, repeated calibration runs, deterministic or frozen reference stats, and backbone-specific calibration files. The stats_sq subset is a dedicated Innovation 3 SQ-style calibration-statistics set and should be used only with the corresponding SQ / DAQ-MPQ scripts. For residual ImageNet experiments, some filenames retain historical prefixes for compatibility with earlier scripts, but they correspond to the final corrected residual-backbone evaluation setting. The results directory stores the final machine-readable outputs used to support the paper tables and quantitative comparisons. It is organized by paper module: innovation1_daq, innovation2_mpq, innovation3_daqmpq, and paper-oriented summary files where applicable. The Innovation 3 directory contains the most important final DAQ-MPQ results, including the main CIFAR-10 MobileNetV2 outputs and the CIFAR / ImageNet residual-backbone generalization results. The figures directory stores exported plots. The main_paper subdirectory contains the figures intended to correspond directly to the final paper, while the supplementary subdirectory contains supporting visualizations used for interpretation and additional analysis. Recommended usage:1. identify the target experiment family2. load the corresponding checkpoint3. use the matching calibration statistics4. verify the final JSON / CSV results5. compare the exported figures against the paper This asset package is intended to be used together with the DAQ-MPQ code repository.