Public articles linked to the same research event.
arXiv This work frames ALMA bandpass calibration anomaly identification as supervised classification, using XGBoost over features built from Nadaraya-Watson kernel-regression scan statistics plus expert-informed features to learn non-linear decision boundaries; evaluated on 42,286 polarization-pair samples from Cycle 9 with execution-block-level nested cross-validation, it reduces both the false-negative and false-positive rates relative to the current ALMA pipeline subband heuristic on 19,567 applicable FDM polarization-pair samples, and shows consistent generalization on roughly an order-of-magnitude larger Cycle 11 dataset including the new Band 1.
This work frames ALMA bandpass calibration anomaly identification as supervised classification, using XGBoost over features built from Nadaraya-Watson kernel-regression scan statistics plus expert-informed features to learn non-linear decision boundaries; evaluated on 42,286 polarization-pair samples from Cycle 9 with execution-block-level nested cross-validation, it reduces both the false-negative and false-positive rates relative to the current ALMA pipeline subband heuristic on 19,567 applicable FDM polarization-pair samples, and shows consistent generalization on roughly an order-of-magnitude larger Cycle 11 dataset including the new Band 1.
This work frames ALMA bandpass calibration anomaly identification as supervised classification, using XGBoost over features built from Nadaraya-Watson kernel-regression scan statistics plus expert-informed features to learn non-linear decision boundaries; evaluated on 42,286 polarization-pair samples from Cycle 9 with execution-block-level nested cross-validation, it reduces both the false-negative and false-positive rates relative to the current ALMA pipeline subband heuristic on 19,567 applicable FDM polarization-pair samples, and shows consistent generalization on roughly an order-of-magnitude larger Cycle 11 dataset including the new Band 1.
This work frames ALMA bandpass calibration anomaly identification as supervised classification, using XGBoost over features built from Nadaraya-Watson kernel-regression scan statistics plus expert-informed features to learn non-linear decision boundaries; evaluated on 42,286 polarization-pair samples from Cycle 9 with execution-block-level nested cross-validation, it reduces both the false-negative and false-positive rates relative to the current ALMA pipeline subband heuristic on 19,567 applicable FDM polarization-pair samples, and shows consistent generalization on roughly an order-of-magnitude larger Cycle 11 dataset including the new Band 1.