From data to noise to data for mixing physics across temperatures with generative artificial intelligence

kntkb / technotes

Personal technical notes

1 stars 0 forks source link

From data to noise to data for mixing physics across temperatures with generative artificial intelligence #21

Open kntkb opened 1 year ago

kntkb commented 1 year ago

PNAS arxiv

kntkb commented 1 year ago

任意の温度のシミュレーション結果を得るため、レプリカ交換モンテカルロ法で得た複数温度のシミュレーション結果を配置と分子毎の運動エネルギーの同時確率に基づくサンプリングとみなし、拡散モデルを使ってそれを学習。学習外の低温度も外挿可能 referenced from twitter by Daisuke Okunohara