near-optimal batch mode active learning and adaptive
TRANSCRIPT
![Page 1: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/1.jpg)
Near-optimal Batch Mode Active Learning and Adaptive Submodular Optimization
Yuxin Chen Andreas KrauseDepartment of Computer Science, ETH Zurich
![Page 2: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/2.jpg)
Batch Mode Active Learning
w
?
![Page 3: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/3.jpg)
![Page 4: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/4.jpg)
?
![Page 5: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/5.jpg)
?
![Page 6: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/6.jpg)
?
![Page 7: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/7.jpg)
?
![Page 8: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/8.jpg)
?
![Page 9: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/9.jpg)
?
![Page 10: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/10.jpg)
? ?
. . .
. . .
![Page 11: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/11.jpg)
?
?
. . .
. . .
![Page 12: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/12.jpg)
?
. . .
?
![Page 13: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/13.jpg)
Multi-stage Influence Maximization in Social Networks
![Page 14: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/14.jpg)
How should we construct the batches?
![Page 15: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/15.jpg)
B
A
s
. . .
![Page 16: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/16.jpg)
THE BatchGreedy ALGORITHM
B
A
s
. . .
Conditional marginal benefit of an item s:
Expectation over all realizations
within the batch
Conditioning on previous
observations
�f (s | A,yB) = EyV
⇥f(y{s}[A[B)� f(yA[B) | yB
⇤.
![Page 17: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/17.jpg)
THE BatchGreedy ALGORITHM
B
A
s
. . .
si,j = argmax
s2V�f (s | {s1,j , . . . , si�1,j}| {z }
the jth batch A
,yB)
BatchGreedy will greedily select the i-th element in the j-th batch
![Page 18: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/18.jpg)
BatchGreedy VS. OPTIMAL BATCH
Cost of BatchGreedy Cost of optimal batch policy
. . .
. . . . . .
�����
�����
�����
�����
![Page 19: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/19.jpg)
BatchGreedy VS. OPTIMAL BATCH
Cost of BatchGreedy Cost of optimal batch policy
. . .
. . . . . .
�����
�����
�����
����� O (lnQ) ·
![Page 20: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/20.jpg)
How many extra items will we select?
![Page 21: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/21.jpg)
BatchGreedy VS. SEQUENTIAL
. . .
Cost of BatchGreedy Cost of optimal sequential policy
. . .
. . .
�����
�����
�����
�����
![Page 22: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/22.jpg)
BatchGreedy VS. SEQUENTIAL
. . .
Cost of BatchGreedy Cost of optimal sequential policy
. . .
. . .
�����
�����
�����
�����
COMPETITIVE
![Page 23: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/23.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
![Page 24: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/24.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
![Page 25: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/25.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
random
![Page 26: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/26.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
sequential
![Page 27: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/27.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
![Page 28: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/28.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
![Page 29: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/29.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
![Page 30: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/30.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
% it
em n
ot c
over
ed
Number of items selected
![Page 31: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/31.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
% it
em n
ot c
over
ed
Number of items selected
0 200 400 600 8000
5
10
15
20
25
30
35
Non−adaptive
![Page 32: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/32.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
% it
em n
ot c
over
ed
Number of items selected
0 200 400 600 8000
5
10
15
20
25
30
35
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
Non−adaptive
![Page 33: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/33.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
% it
em n
ot c
over
ed
Number of items selected
0 200 400 600 8000
5
10
15
20
25
30
35
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
10−batch
0 200 400 600 8000
5
10
15
20
25
30
35
Non−adaptive
![Page 34: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/34.jpg)
EXPERIMENTAL RESULTS
# POSTER
w
?
. . .
. . .
10 20 30 40 500
2
4
6
8
10
12
14
Number of labels requested
% M
ista
kes
10 20 30 40 500
2
4
6
8
10
12
14
10−batch greedy
10 20 30 40 500
2
4
6
8
10
12
14
random
10 20 30 40 500
2
4
6
8
10
12
14
KLR−BMAL
10 20 30 40 500
2
4
6
8
10
12
14
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
% it
em n
ot c
over
ed
Number of items selected
0 200 400 600 8000
5
10
15
20
25
30
35
sequential
0 200 400 600 8000
5
10
15
20
25
30
35
10−batch
0 200 400 600 8000
5
10
15
20
25
30
35
100−batch
0 200 400 600 8000
5
10
15
20
25
30
35
Non−adaptive
![Page 35: Near-optimal Batch Mode Active Learning and Adaptive](https://reader033.vdocuments.us/reader033/viewer/2022061102/629cf53408fad228b14c49e0/html5/thumbnails/35.jpg)
Code Available
www.inf.ethz.ch/~chenyux/icml13/bmal-src.zip
Thanks for your attention
Come to see our poster!