A community template on using AI in Quantum Chemistry applications #157
Conversation
|
Hi, thanks for the contribution! Would it be possible to avoid adding the many template-specific files directly but add instructions on how to download them in a README in the template folder? Cheers! |
|
Thanks for the suggestions. I've moved the datasets (they're tiny) to a different GitHub repository and explained how to download them, along with all the other required steps, in a README file. Let me know if there's anything else to improve. |
templates/MACE/ideas.json
Outdated
| @@ -0,0 +1,38 @@ | |||
| [ | |||
| { | |||
| "Name": "number of interactions", | |||
There was a problem hiding this comment.
Names should have _ in them and no spaces!
There was a problem hiding this comment.
somehow WSL2 produced them, deleted now
templates/MACE/latex/template.tex
Outdated
|
|
||
| @inproceedings{ | ||
| edm, | ||
| title={Elucidating the Design Space of Diffusion-Based Generative Models}, |
There was a problem hiding this comment.
inappropriate base cites for this paper!
|
Yes, indeed, there were erroneous files. I've deleted them and renamed the ideas. The LaTeX template was mistakenly taken from a diffusion experiment without any changes to the base references. I was running my experiments using the incorrect template. Funny enough, all of the citations in the final text of the "papers" produced were correct, except for one citing a review on diffusion models. I reran the experiment with a corrected LaTeX template and updated the link to the example paper in README.md. Sorry for so late. Lunar New Year.... |
|
Tysm! |
This template provides a slightly non-computer-science example. One of the important topics in modern quantum chemistry is the acceleration of computations. To this end, AI techniques, such as the MACE package implementing neural network potentials, are used for calculating energies and forces.
While MACE-type equivariant neural networks excel in prediction accuracy, their application to molecular dynamics simulations reveals a less understood connection between the specific architecture of the MACE family and the success of molecular dynamics simulations. This slightly, but not fully, toy example sets the stage for an open-ended study.
An example of a generated paper can be found https://drive.google.com/file/d/1dO8qXeCWaXkqP-sbV4z1S5RwY--pmggh/view?usp=sharing
Three extra packages are required:
pip install mace-torchpip install MDAnalysispip install statsmodels