This is the code for our paper Privacy-Aware Randomized Quantization via Linear Programming.

To obtain the results shown in Figure 3, run compare_err_uniform.py. For instance:

python compare_err_uniform.py --c 1 --eps 1

To obtain the results shown in Table 2, run compare_err_gauss.py.

To obtain the results shown in Figure 4(a), run vec_err.py:

python vec_err.py --c 1 --type l1 --dim 10

To obtain the results shown in Figure 4(b), run vec_err.py:

python vec_err.py --c 1 --type l2 --dim 100

To obtain the results shown in Figure 4(c), run dp_sgd_cancer.py:

python dp_sgd_cancer.py --c 0.1 --eps 1

To obtain the results shown in Figure 4(d), run dp_sgd_mnist.py:

python dp_sgd_cancer.py --c 0.01 --eps 1

The results of MVU mechanism can be recorded from dp_compression project.