tensorflow / tensorflow/privacy
The current epsilon is very large in mnist_dpsgd_tutorial_keras.py
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 2k
- Forks
- 477
- Avg merge
- 22h 12m
- Merged PRs (30d)
- 1
Description
I run the mnist_dpsgd_tutorial.py with noise_multiplier=1.1, l2_norm_clip=1.0, batch_size=250, epochs=2, and get the current eps 0.8
But when I run mnist_dpsgd_tutorial_keras.py with the same parameters, the current eps is 16890.31.
Why is there such a big gap in their privacy budgets?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by running tutorials/mnist_dpsgd_tutorial.py and tutorials/mnist_dpsgd_tutorial_keras.py with noise_multiplier=1.1, l2_norm_clip=1.0, batch_size=250, and epochs=2. Compare how each tutorial computes its privacy budget. Done means the discrepancy is explained and any confirmed accounting inconsistency is addressed.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning, security
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 32/100