@cedergroup : 💡Why does fine-tuning work well with pre-trained universal MLIP? We benchmarked several uMLIPs and discovered a systematic PES softening effect, which suggests that a considerable fraction of errors can be efficiently corrected. Check the preprint. 👇 arxiv.org/abs/24505.07105 • TwiCopy

Ceder Group

@cedergroup

+ Follow

Computational and Experimental Design of Emerging materials Research (CEDER) Group at @BerkeleyMSE & @BerkeleyLab

ID: 743114133294317568

linkhttp://ceder.berkeley.edu calendar_today15-06-2016 16:13:10

157 Tweet

2,2K Takipçi

135 Takip Edilen

Ceder Group

@cedergroup

a year ago

💡Why does fine-tuning work well with pre-trained universal MLIP? We benchmarked several uMLIPs and discovered a systematic PES softening effect, which suggests that a considerable fraction of errors can be efficiently corrected. Check the preprint. 👇 arxiv.org/abs/2405.07105

thumb_up_off_alt36

chat_bubble_outline0

repeat9

shareShare