Bi-objective Continual Learning: Learning 'New' while Consolidating 'Known'
Comparison Results
The detailed test accuracies of each method under different datasets are listed in the following table. The reported results are avearaged over 5 runs with different settings of patterns.
CIFAR10 with the full set of training samples
Method/Sessions
1
2
3
4
5
Naive
67.80
69.31
71.37
73.12
73.23
Cumulative
67.80
76.13
81.22
81.73
82.12
EWC*
67.80
69.45
72.68
74.02
74.31
SI*
67.80
70.48
72.82
74.63
74.58
IMM*
67.80
69.69
72.85
74.37
73.84
EEIL*
67.80
71.97
73.27
74.91
74.66
A-GEM*
67.80
72.27
73.72
74.81
75.15
Our-PLC
67.80
72.75
75.16
76.01
76.98
Our-PLC-CPL
67.80
73.60
76.01
76.65
77.43
CIFAR10 with the randomly sampled set of training samples
Method/Sessions
1
2
3
4
5
Naive
67.80
67.61
67.86
68.40
68.19
Cumulative
67.80
71.82
72.30
72.60
72.72
EWC*
67.80
68.27
69.32
68.99
69.27
SI*
67.80
68.56
69.53
69.27
70.24
IMM*
67.80
68.07
69.65
69.95
70.38
EEIL*
67.80
69.83
69.01
70.01
69.96
A-GEM*
67.80
70.28
70.71
69.97
70.97
Our-PLC
67.80
70.10
70.79
70.69
71.27
Our-PLC-CPL
67.80
71.50
71.90
71.70
72.10
CIFAR100 with the full set of training samples
Method/Sessions
1
2
3
4
5
Naive
39.73
44.00
46.28
47.59
49.16
Cumulative
39.73
47.60
51.10
55.19
59.03
EWC*
39.73
44.52
46.88
48.17
49.83
SI*
39.73
44.61
46.95
48.68
50.92
IMM*
39.73
44.87
47.03
49.38
50.98
EEIL*
39.73
45.03
47.65
50.22
51.12
A-GEM*
39.73
45.13
47.89
50.21
51.24
Our-PLC
39.73
46.46
49.11
51.12
52.90
Our-PLC-CPL
39.73
47.15
49.78
51.64
53.76
CIFAR100 wih the randomly sampled set of training samples
Method/Sessions
1
2
3
4
5
Naive
39.73
37.43
37.03
37.29
36.77
Cumulative
39.73
42.32
43.83
45.34
46.75
EWC*
39.73
37.87
37.85
38.26
38.28
SI*
39.73
38.76
39.07
39.48
39.81
IMM*
39.73
39.74
39.89
40.21
40.25
EEIL*
39.73
39.73
39.09
39.62
40.01
A-GEM*
39.73
40.01
40.19
40.68
40.89
Our-PLC
39.73
40.95
41.00
41.12
41.55
Our-PLC-CPL
39.73
40.94
41.07
41.21
41.87
CORe50 with the full set of training samples
Method/Sessions
1
2
3
4
5
6
7
8
Naive
51.00
64.50
60.10
62.41
59.97
63.26
65.09
67.61
Cumulative
51.00
65.10
72.30
74.80
77.20
77.80
78.50
79.20
EWC*
51.00
63.98
61.02
62.72
59.84
63.66
65.12
66.89
SI*
51.00
64.01
62.15
61.56
59.48
63.88
64.85
67.21
IMM*
51.00
64.51
62.51
63.21
60.12
63.32
66.12
68.54
EEIL*
51.00
63.74
64.88
62.56
60.56
64.45
67.00
70.00
A-GEM*
51.00
64.72
65.17
63.54
61.54
66.48
67.21
71.54
Our-PLC
51.00
63.46
65.76
64.49
64.04
68.65
70.20
73.15
Our-PLC-CPL
51.00
63.52
66.86
64.92
64.58
69.67
71.14
74.31
CORe50 with the randomly sampled set of training samples
Method/Sessions
1
2
3
4
5
6
7
8
Naive
51.00
57.00
50.69
55.26
49.81
59.94
60.52
63.00
Cumulative
51.00
60.76
67.14
70.41
72.59
74.20
74.90
76.10
EWC*
51.00
57.73
52.99
56.34
52.41
60.40
61.84
64.53
SI*
51.00
57.81
53.06
56.46
52.64
60.53
61.91
64.73
IMM*
51.00
57.96
53.58
56.61
53.23
60.61
62.04
64.95
EEIL*
51.00
57.12
55.95
56.72
55.21
60.84
62.67
65.12
A-GEM*
51.00
58.03
56.16
57.65
55.52
60.76
63.11
65.24
Our-PLC
51.00
59.80
58.35
58.75
58.91
61.15
65.19
66.98
Our-PLC-CPL
51.00
60.11
58.77
59.05
59.03
62.68
65.39
67.83
SubImageNet with the full set of training samples
Method/Sessions
1
2
3
4
5
Naive
22.06
28.49
34.21
40.48
45.46
Cumulative
22.06
31.36
42.29
52.98
58.27
EWC*
22.06
28.32
34.28
38.26
40.73
SI*
22.06
28.61
34.57
40.14
47.86
IMM*
22.06
29.28
35.64
42.64
48.52
EEIL*
22.06
28.45
34.08
41.05
47.84
A-GEM*
22.06
29.39
35.84
42.38
47.17
Our-PLC
22.06
29.59
36.38
43.19
49.16
Our-PLC-CPL
22.06
30.15
37.32
44.51
50.32
SubImageNet with the randomly sampled set of training samples
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent