Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

MI between the model and training dataset #1

Open
ayhem18 opened this issue Mar 27, 2024 · 0 comments
Open

MI between the model and training dataset #1

ayhem18 opened this issue Mar 27, 2024 · 0 comments

Comments

@ayhem18
Copy link

ayhem18 commented Mar 27, 2024

First of all, thank you for sharing such an influential work with the public.
The findings presented in your work might represent theoretical grounds for my empirical results. Thus, I have a couple of questions mainly concerning the calculation of MI between the model and the training dataset:
image

if I am not mistaken this quantity is computed in the "compute_MI_theta_D_single_seed_jensen" function found in the following file

image

The 'data_instances' argument in the screenshot above is only used as 'list(range(5))'. Does that mean that you are using 1 copy of the 'swag' model to compute the first term in the equation above: $$\log p(w^j|s)$$. The value would be stored in the 'log_posterior' variable

and 4 copies of the same model to estimate $$(\frac{1}{|D|} \sum_{s^{'} \in D} \log p(w^j| s^{'})$$. The value would be stored in the '
log_prior' variable.

if the models represented by the data instances are not the same, would you please highlight the difference and indicate the part of the code where this difference is implemented.

Thanks a lot in advance.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant