MIT researchers develop data-efficient training technique for climate emulators

MIT researchers have developed a training technique that lets AI climate emulators match or outperform conventionally trained models while using only a fraction of the data. The method generates optimized synthetic training scenarios with rapid emissions variations, so an emulator trained on far less data can accurately predict unseen scenarios where traditional emulators failed. The work is published in the Proceedings of the National Academy of Sciences. The researchers say the approach also improves the emulators' adaptability, though the reported gains come from their own study rather than independent evaluation.

Generated from reports · updated 1 h ago

Not enough continuous observations yet to draw a trend.