MENO learns only nonlinear plasma simulation parts

Jorge Bravo Abad posted that the idea behind MENO is to avoid asking neural networks to relearn known physics. In plasma simulations most dynamics follow essentially linear equations, so the model learns only the small strongly nonlinear fraction while applying the known mathematics to the rest. The tweet includes an attached scientific figure. Bharath Ramsundar retweeted the post from the director of the AI for Materials Lab at UAM Madrid.