For the past few months, I've been studying the basics of diffusion generative model, for which Jianlin Su has written a series of blogs in Chinese, which I found much easier to understand the logics behind. So I've decided to translate it, and here are the first 11 blogs.
@TimothyDuignan Just had a try on MACE foundation model, interesting results although not something unexpected. Materials that are more stable mechanicallly did show better performance compared to those that are not stable. Will do more analyses and probably write a small paper on it:-)
@cryptopatrick@Jianlin_S I'm glad that someone else also found it useful, not just me killing my time. :-)
Unfortunately it's a cold wetty winter down here in Australia. 😪
For the past few months, I've been studying the basics of diffusion generative model, for which Jianlin Su has written a series of blogs in Chinese, which I found much easier to understand the logics behind. So I've decided to translate it, and here are the first 11 blogs.
@ChengBingqing Just saw your post. This is very interesting as I have collaborators who works a lot in ferroelectrics, it opens up lots of possibilities with MLP.
In scientific computation, there's a large cultural divide between physics- and ML-based approaches.
My latest blog post argues that physics–ML hybrids will be the future, despite offending both fields. (ft. spicy quotes from @pranamanam + some wisdom from @manntis4)
After spending last couple of months reading on diffusion generative models for materials discoveries, I'm wondering are we being taken over by the mathematicians? Surely the maths & models are cool & hard, but I failed to get much inspiration in materials science perspective.😅
I think, if I am correct, generative AI for crystal generation is a bit of an overkill, a simple mathematical optimisation under constrains may work well enough, as shown here.
Utterly not impressed with the reviewer from @PCCP, after two rounds of reviews and for the first time, I am considering withdrawing my submission from the journal. I rather just leave it on @ChemRxiv.
Microsoft researchers introduce MatterGen, a model that can discover new materials tailored to specific needs—like efficient solar cells or CO2 recycling—advancing progress beyond trial-and-error experiments. https://t.co/z9yOaV7VGo