privacy
privacy copied to clipboard
tensorflow
Library for training machine learning models with privacy for training data
When running the `mnist_dpsgd_tutorial_keras.py` with all default parameters, the train/test accuracies do not improve over the course of training. The final accuracy is around 10%, which is of course equivalent...
### Set-up: tensorflow==2.4.0 tensorflow-privacy == 0.5.1 plaidml == 0.7.0 plaidml-keras == 0.7.0 Mac OS 10.15.7 (19H114) 16" AMD Radeon Pro 5300M 4 GB ### Problem: Hello everyone. I have been...
Hello! I have a question about Gradient Clipping, that arises from the following principles of privacy accounting and DP-SGD: The RDP calculation for each step in training is based on...
Hi, first of all, thanks for continuing to expand and improve this amazing repo. I've noticed a small issue with the usage of the new `DPKeras*` optimizers (the non vectorized...
In [mnist_lr_tutorial.py](https://github.com/tensorflow/privacy/blob/1860ee1/tutorials/mnist_lr_tutorial.py), the computation of the DP budget for the DP-SGD algorithm (here: [line 177](https://github.com/tensorflow/privacy/blob/1860ee1/tutorials/mnist_lr_tutorial.py#L177)) is missing a scaling of `noise_multiplier` by `batch_size` to match the computation for the amplification-by-iteration...
I am a little confused about how rdp_accountant.py works. Could you please provide the papers or formula you used in rdp_accountant.py especially about the method to compute rdp? That will...
serialise dp parameters
Hi, I encountered a problem when adding regularization to Keras models using the new DPKerasOptimizer's with TF2.4.0 with (Sparse)CategoricalCrossentropy. The error (see below) can be reproduced by adding `kernel_regularizer=tf.keras.regularizers.l2(1e-8)`to a...
The vectorized optimizer in [`mnist_dpsgd_tutorial_vectorized.py`](https://github.com/tensorflow/privacy/blob/master/tutorials/mnist_dpsgd_tutorial_vectorized.py) runs significantly faster than the new TF2 Keras optimizer used in [`mnist_dpsgd_tutorial_keras.py`](https://github.com/tensorflow/privacy/blob/master/tutorials/mnist_dpsgd_tutorial_keras.py). On a Titan RTX (24 GB VRAM) at batch size 250 and 250...
Metadata
Owner
Metadata
Library for training machine learning models with privacy for training data