Dr. Kaiming Cui

A comprehensive analysis of main-sequence stars with rotation periods exceeding 30 days using Kepler SAP light curves. We developed a systematic approach combining light curve pre-processing, advanced period detection algorithms, and rigorous candidate selection criteria. Our method successfully identified over 1000 long-period rotators, including 165 newly discovered systems, significantly expanding our understanding of stellar rotation in the slow-rotation regime.

Comprehensive photometric, spectroscopic, asteroseismic, and evolutionary analysis of the Algol-type eclipsing binary KIC 12268220. Our study reveals a complex system containing a δ Scuti pulsating star and an evolved low-mass secondary (~0.23 M☉). Evolutionary modeling suggests the secondary is in a protohelium white dwarf phase, representing a transitional state similar to R CMa-type systems that may evolve into EL CVn binaries.

Development of a novel convolutional neural network architecture for automated detection of planetary transits in photometric time series. Our method, based on proven computer vision frameworks, achieves 94% precision and 95% recall for transits with signal-to-noise ratios above 6. The algorithm significantly reduces false positive rates compared to traditional detection methods and has been successfully applied to Kepler and TESS datasets.

Large-scale statistical analysis of substellar companions using multi-epoch radial velocity, Gaia astrometry, and direct imaging data for 5,108 AFGKM stars. Our novel analysis pipeline identified 167 cold gas giants and 68 brown dwarf companions, revealing a pronounced brown dwarf desert around 40 Jupiter masses. The study demonstrates that binary fraction follows a power-law distribution beyond 3 AU, similar to single stars, indicating no multiplicity preference for brown dwarf formation.

Novel deep learning framework for automated classification of astronomical light curves using weakly supervised object detection. Our model automatically identifies optimal analysis windows in both time and frequency domains, enabling feature extraction across multiple scales and sampling intervals. Trained on diverse datasets from space-based and ground-based observations, the framework achieves 87% accuracy for variable stars and transient classification. The model demonstrates excellent transferability across different surveys without requiring retraining.

Development of a novel convolutional neural network architecture for automated detection of planetary transits in photometric time series. Our method, based on proven computer vision frameworks, achieves 94% precision and 95% recall for transits with signal-to-noise ratios above 6. The algorithm significantly reduces false positive rates compared to traditional detection methods and has been successfully applied to Kepler and TESS datasets.