update tutorial README

PiperOrigin-RevId: 237161576
This commit is contained in:
Nicolas Papernot 2019-03-06 18:37:13 -08:00 committed by A. Unique TensorFlower
parent fdd110b7df
commit d226cf3172

View file

@ -56,8 +56,9 @@ compute the privacy guarantee) are:
## Expected Output ## Expected Output
When the script is run with the default parameters, the output will When the `mnist_dpsgd_tutorial.py` script is run with the default parameters,
contain the following lines (leaving out a lot of diagnostic info): the output will contain the following lines (leaving out a lot of diagnostic
info):
``` ```
... ...
Test accuracy after 1 epochs is: 0.774 Test accuracy after 1 epochs is: 0.774
@ -81,14 +82,18 @@ compute_dp_sgd_privacy.py --N=60000 --batch_size=256 --noise_multiplier=1.1 --ep
``` ```
allows us to conclude, in a matter of seconds, that DP-SGD run with default allows us to conclude, in a matter of seconds, that DP-SGD run with default
parameters satisfies differential privacy with eps = 3.01 and delta = 1e-05. parameters satisfies differential privacy with eps = 3.01 and delta = 1e-05.
Note that the flags provided in the command above correspond to the tutorial in
`mnist_dpsgd_tutorial.py`. The command is applicable to other datasets but the
values passed must be adapted (e.g., N the number of training points).
## Select Parameters ## Select Parameters
The table below has a few sample parameters illustrating various accuracy/privacy The table below has a few sample parameters illustrating various
tradeoffs (default parameters are in __bold__; privacy epsilon is reported accuracy/privacy tradeoffs achieved by the MNIST tutorial in
at delta=1e-5; accuracy is averaged over 10 runs, its standard deviation is `mnist_dpsgd_tutorial.py` (default parameters are in __bold__; privacy epsilon
less than .3% in all cases). is reported at delta=1e-5; accuracy is averaged over 10 runs, its standard
deviation is less than .3% in all cases).
| Learning rate | Noise multiplier | Clipping threshold | Number of microbatches | Number of epochs | Privacy eps | Accuracy | | Learning rate | Noise multiplier | Clipping threshold | Number of microbatches | Number of epochs | Privacy eps | Accuracy |
| ------------- | ---------------- | ----------------- | ---------------------- | ---------------- | ----------- | -------- | | ------------- | ---------------- | ----------------- | ---------------------- | ---------------- | ----------- | -------- |