query_obfuscation.py

import warnings
import matplotlib.pyplot as plt
from sklearn.datasets import fetch_openml
from sklearn.exceptions import ConvergenceWarning
from sklearn.model_selection import train_test_split
from sklearn.neural_network import MLPClassifier
import numpy as np

# Load data from https://www.openml.org/d/554
X, y = fetch_openml("mnist_784", version=1, return_X_y=True, as_frame=False)
X = X / 255.0

# Split data into train partition and test partition
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0, test_size=0.7)

mlp = MLPClassifier(
    hidden_layer_sizes=(40,),
    max_iter=8,
    alpha=1e-4,
    solver="sgd",
    verbose=10,
    random_state=1,
    learning_rate_init=0.2,
)

with warnings.catch_warnings():
    warnings.filterwarnings("ignore", category=ConvergenceWarning, module="sklearn")
    mlp.fit(X_train, y_train)

print("Training set score: %f" % mlp.score(X_train, y_train))
print("Test set score: %f" % mlp.score(X_test, y_test))

noises = [0, 0.2, 0.4, 0.6, 0.8, 1]
scores = []
baselines = []
num_tests = 1000
num_nudges = 15
for noise_level in noises:
    correct = 0
    baseline = 0
    for i in range(num_tests):
        # Baseline noisy data
        noisy_data = X_test[i] + np.random.uniform(-noise_level,noise_level, (28*28,))
        np.clip(noisy_data,0,1)
        bs_probs =mlp.predict_proba([noisy_data])
        if (int(np.argmax(bs_probs)) == int(y_test[i])): baseline += 1

        # Linear approximation
        probs = [0 for _ in range(10)]
        for j in range(num_nudges):
            noisy_data = X_test[i] + np.random.uniform(-noise_level,noise_level, (28*28,))
            np.clip(noisy_data,0,1)
            probs += mlp.predict_proba([noisy_data])/num_nudges
        if (int(np.argmax(probs)) == int(y_test[i])): correct += 1

    b_score = baseline / num_tests
    baselines.append(b_score)

    score = correct / num_tests
    scores.append(score)

print("Noisy set score: " + str(score))

fig, (ax1, ax2, ax3) = plt.subplots(1,3)

ax1.matshow((X_test[0] + np.random.normal(0,1, (28*28,))).reshape(28, 28), cmap=plt.cm.gray, vmin=0, vmax=1)

ax2.plot(noises, baselines, color="red")
ax2.set_title("Baseline Noisy Data Accuracy")
ax2.set_xlabel("Noise Level")
ax2.set_ylabel("Classification Accuracy")
ax2.set_ylim([0.7, 1])

ax3.plot(noises, scores, color="green")
ax3.set_title("Obfuscated Query Accuracy")
ax3.set_xlabel("Noise Level")
ax3.set_ylabel("Classification Accuracy")
ax3.set_ylim([0.8, 1])

plt.show()