X-DeepSCA/ANN_SCA_Hyperparam_analysis.py at master · SparcLab/X-DeepSCA · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
import tensorflow as tf
import numpy as np
import matplotlib
from random import sample
import scipy.io as sio
matplotlib.use('WXAgg')
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D


# Data stored in .mat format
# two fields
#   traces - array of traces
#   key - array of key used

# devices G2 - G5 are used for training
g2 = sio.loadmat(".\\mat_traces\\cw308XGD2_10k_nov5_1447.mat")
g2_traces = g2['traces'][:, 0:500]
g2_keys = g2['key'][:, 0]

g3 = sio.loadmat(".\\mat_traces\\cw308XGD3_10k_nov5_1643.mat")
g3_traces = g3['traces'][:, 0:500]
g3_keys = g3['key'][:, 0]

g4 = sio.loadmat(".\\mat_traces\\cw308XGD4_10k_nov8_2228.mat")
g4_traces = g4['traces'][:, 0:500]
g4_keys = g4['key'][:, 0]

g5 = sio.loadmat(".\\mat_traces\\cw308XGD5_10k_nov9_1538.mat")
g5_traces = g5['traces'][:, 0:500]
g5_keys = g5['key'][:, 0]

# Devices g6-g9 used for testing cross-device accuracy
g6 = sio.loadmat(".\\mat_traces\\cw308XGD6_10k_nov9_1559.mat")
g6_traces = g6['traces'][:, 0:500]
g6_keys = g6['key'][:, 0]

g8 = sio.loadmat(".\\mat_traces\\cw308XGD8_50k_nov14_1635.mat")
g8_traces = g8['traces'][:, 0:500]
g8_keys = g8['key'][:, 0]

g9 = sio.loadmat(".\\mat_traces\\cw308XGD9_nov14_2011.mat")
g9_traces = g9['traces'][:, 0:500]
g9_keys = g9['key'][:, 0]

# g2 = sio.loadmat(".\\mat_traces\\cw308XGD2_10k_nov5_1447.mat")
# g2_traces = g2['traces'][:, 0:500]
# g2_traces = g2['key'][:, 0]

# Concantenate device traces and keys to get training and testing sets

train_traces = np.vstack([g2_traces, g3_traces, g4_traces, g5_traces])
train_keys = np.hstack([g2_keys, g3_keys, g4_keys, g5_keys])

test_traces = np.vstack([g6_traces, g8_traces, g9_traces])
test_keys = np.hstack([g6_keys, g8_keys, g9_keys])


N = train_traces.shape[0]
train_indices = sample(range(N), int(0.9 * N))
x_train = train_traces[train_indices]
x_test = np.delete(train_traces, train_indices, 0)
y_train = train_keys[train_indices]
y_test = np.delete(train_keys, train_indices, 0)


# samplewise standardization seems mose effective - 6-8% acc on device 2


x_train = (x_train - np.mean(x_train, axis=0)) / (np.std(x_train, axis=0))
x_test = (x_test - np.mean(train_traces[train_indices], axis=0)) / (np.std(train_traces[train_indices], axis=0))
test_traces = (test_traces - np.mean(train_traces[train_indices], axis=0)) / \
    (np.std(train_traces[train_indices], axis=0))

# Hyperparameter sweeps: example of batch size
test_acc = []
for bs in np.arange(16, 1024, 32):
    # Fully Connected Network model
    model_dense = tf.keras.models.Sequential([
        tf.keras.layers.Reshape((500, ), input_shape=(500, )),
        tf.keras.layers.Dense(200, activation=tf.nn.relu,
                              kernel_regularizer=tf.keras.regularizers.l2(0)),
        tf.keras.layers.BatchNormalization(),
        tf.keras.layers.Dropout(0.1),
        tf.keras.layers.Dense(200, activation=tf.nn.relu,
                              kernel_regularizer=tf.keras.regularizers.l2(0)),
        tf.keras.layers.BatchNormalization(),
        tf.keras.layers.Dropout(0.05),
        tf.keras.layers.Dense(256, activation=tf.nn.softmax)
    ])
    model = model_dense
    # Use normal Adam optimizer, default parameters
    model.compile(optimizer="adam",
                  loss='sparse_categorical_crossentropy',
                  metrics=['accuracy'])
    model.summary()
    # Train the model for 10 epochs
    history = model.fit(x_train, y_train, validation_split=0.11,
                        epochs=10, shuffle=True, batch_size=bs)
    # Evaluate same device performance
    metrics = model.evaluate(x_test, y_test)
    # Evaluate cross device performance
    test_metrics = model.evaluate(test_traces, test_keys)
    print(metrics)
    print(test_metrics)
    # save cross device accuracy
    test_acc.append(test_metrics[1])

# Save accuracy results for further analysis
np.save("cross_dev_batch_size_accs.npy", test_acc)

# Plot results here if wanted
plt.rc('font', family='serif', weight='bold')
plt.rc('xtick', labelsize='x-small')
plt.rc('ytick', labelsize='x-small')
fig = plt.figure(figsize=(4, 3))
ax = fig.add_subplot(1, 1, 1)

x = [r for r in range(16, 1024, 32)]
plt.plot(x, test_acc, linewidth=2)
plt.title('Model Accuracy vs Batch Size on G6')
ax.set_ylabel('Accuracy (%)', fontweight='bold')
ax.set_xlabel('Batch Size', fontweight='bold')
plt.show()