AI RESEARCH
Do Masked Autoencoders Improve Downhole Prediction? An Empirical Study on Real Well Drilling Data
arXiv CS.LG
•
ArXi:2604.20909v1 Announce Type: new Downhole drilling telemetry presents a fundamental labeling asymmetry: surface sensor data are generated continuously at 1~Hz, while labeled downhole measurements are costly, intermittent, and scarce. Current machine learning approaches for downhole metric prediction universally adopt fully supervised