-
Notifications
You must be signed in to change notification settings - Fork 0
/
results4-but-1-layer-remaining.txt
107 lines (84 loc) · 5.28 KB
/
results4-but-1-layer-remaining.txt
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
with # samples 500 we got best acc of 10.0, batch of 128, lr = 0.05 5 layere remaining cu 50 epochs
with # samples 2500 we got best acc of 50.4, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 57.25, batch of 128, lr = 0.05
inainte am avut cu 50 epochs si acum fac cu 100
with # samples 500 we got best acc of 12.37, batch of 128, lr = 0.05 5 layere remaining
with # samples 2500 we got best acc of 56.83, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 61.67, batch of 128, lr = 0.05
100 epochs si 4 layere remainig
with # samples 500 we got best acc of 26.06, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 55.9, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 60.4, batch of 128, lr = 0.05
3 layers remaining 100 epochs
with # samples 500 we got best acc of 42.08, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 52.19, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 53.9, batch of 128, lr = 0.05
3 layers again and run only 5k
with # samples 5000 we got best acc of 53.17, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 52.44, batch of 128, lr = 0.05
with # samples 500 we got best acc of 39.31, batch of 128, lr = 0.05
3 layers and modify attack model
with # samples 5000 we got best acc of 54.86, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 52.59, batch of 128, lr = 0.05
with # samples 500 we got best acc of 42.9, batch of 128, lr = 0.05
5 remaining layers and modified attack model
with # samples 5000 we got best acc of 63.77, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 60.37, batch of 128, lr = 0.05
with # samples 500 we got best acc of 47.57, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 44.06, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 43.18, batch of 128, lr = 0.05
with # samples 500 we got best acc of 39.92, batch of 128, lr = 0.05
1 remaining layer cand attacker are access la model
with # samples 5000 we got best acc of 42.69, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 42.57, batch of 128, lr = 0.05
with # samples 500 we got best acc of 40.37, batch of 128, lr = 0.05
1 remaining layer but attacker has access to first layers as well
with # samples 5000 we got best acc of 86.67, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 86.5, batch of 128, lr = 0.05
with # samples 500 we got best acc of 86.36, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 86.67, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 86.5, batch of 128, lr = 0.05
with # samples 500 we got best acc of 86.36, batch of 128, lr = 0.05
2 remaining layers but attacker has access to first layers as well
with # samples 5000 we got best acc of 86.74, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 86.62, batch of 128, lr = 0.05
with # samples 500 we got best acc of 86.11, batch of 128, lr = 0.05
3 layers remaining but the attacker doesnt have access to the first layer
with # samples 5000 we got best acc of 86.47, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 86.61, batch of 128, lr = 0.05
with # samples 500 we got best acc of 86.23, batch of 128, lr = 0.05
4 layers remaining with access to the first layer
with # samples 5000 we got best acc of 86.33, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 85.6, batch of 128, lr = 0.05
with # samples 500 we got best acc of 83.99, batch of 128, lr = 0.05
4 layers remaining but the attacker DOESNT HAVE ACEESS TO FIRST LAYER
with # samples 5000 we got best acc of 61.81, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 59.57, batch of 128, lr = 0.05
with # samples 500 we got best acc of 52.72, batch of 128, lr = 0.05
3 layers and attaker doesnt have access to first layer
with # samples 5000 we got best acc of 56.01, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 54.39, batch of 128, lr = 0.05
with # samples 500 we got best acc of 46.5, batch of 128, lr = 0.05
with # samples 5000 we got best acc of 49.1, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 47.39, batch of 128, lr = 0.05
with # samples 500 we got best acc of 42.21, batch of 128, lr = 0.05
2 layers remaining
with # samples 5000 we got best acc of 49.1, batch of 128, lr = 0.05
with # samples 2500 we got best acc of 47.39, batch of 128, lr = 0.05
with # samples 500 we got best acc of 42.21, batch of 128, lr = 0.05
4 layers remaining cu attack-ul nou
with # samples 5000 we got best acc of 63.4, batch of 128, lr = 0.07
with # samples 2500 we got best acc of 60.38, batch of 128, lr = 0.07
with # samples 500 we got best acc of 53.08, batch of 128, lr = 0.07
3 layers remaining
with # samples 5000 we got best acc of 61.56, batch of 128, lr = 0.07
with # samples 2500 we got best acc of 58.17, batch of 128, lr = 0.07
with # samples 500 we got best acc of 50.63, batch of 128, lr = 0.07
2 layers
with # samples 5000 we got best acc of 55.02, batch of 128, lr = 0.07
with # samples 2500 we got best acc of 52.76, batch of 128, lr = 0.07
with # samples 500 we got best acc of 45.24, batch of 128, lr = 0.07
1 layer
with # samples 5000 we got best acc of 44.19, batch of 128, lr = 0.07
with # samples 2500 we got best acc of 43.12, batch of 128, lr = 0.07
with # samples 500 we got best acc of 40.14, batch of 128, lr = 0.07