What is the difference between score model and confidence model？ about diffdock HOT 4 CLOSED

is large score model or small score model Refers to the https://github.com/gcorso/DiffDock/blob/main/workdir/paper_score_model/best_ema_inference_epoch_model.pt ？and The result of running python -m confidence.confidence_train is https://github.com/gcorso/DiffDock/blob/main/workdir/paper_confidence_model/best_model_epoch75.pt ？

from diffdock.

Correct regarding the location, regarding the difference between the two:

• the score model is used to generated a number of poses via the diffusion process, note that it is called score, not because it predicts a score, but because it predicts the score of the distribution which is the gradient of the log density function (think about it as a force)
• the confidence model instead is trying to determine how good the poses generated are

I hope this helps and I suggest to read the paper for more details!
Gabriele

from diffdock.

Correct regarding the location, regarding the difference between the two:

• the score model is used to generated a number of poses via the diffusion process, note that it is called score, not because it predicts a score, but because it predicts the score of the distribution which is the gradient of the log density function (think about it as a force)
• the confidence model instead is trying to determine how good the poses generated are

I hope this helps and I suggest to read the paper for more details! Gabriele

So the Score Model here is the Score
Diffusion Model?

from diffdock.

Yes in the literature of diffusion models, the model predicting the gradient of the log density of the data (i.e. the score) is called "score model"

from diffdock.