李祈均/人類行為訊號處理 : 跨學科 (醫療、教育、心理)...
Post on 21-Apr-2017
2.575 views
TRANSCRIPT
-
1
(Jeremy)Behavioral Informatics and Interaction Computation Lab (BIIC)
:
()
2016
2016.07.17
-
2
-
3
(BSP)
-
BSP INGREDIENTS
4
()
: +
I. II.
III. IV.
-
5
BSP INGREDIENTS
-
6
. . .
-
7
:
-
8
(20133 13 )
(20135 29 )
-
9
-
10
200/
:
?
-
11
62.5 ():" "
89 ():
-
12
:
-
13
:
-
14
:
:
-frame Dense Points Tracking
TRAJ
MBHxy
Each = A Unit-level (66ms) -length Derived Video features
: Dense Trajectory Fisher-
1
2
3
1
2
Acoustic LLDs
Each : = A Unit-level (200ms)-length Dense Acoustic Features
Functionals
1: {1, 1}1
1:1
2:1
:1
1:
: Dense Unit Acoustic Features
2: {1, 2}
3: {1, 3}
4: {1, 4}
K-Means Bag-of-word
-
15
:
late fusion technique
Support vector regression
Support vector regression
+
-
16
:
1
2
Spearman correlation()
= .
-
17
:
2
1
2
2
1
10
= .
= .
= .
-
18
?
-
19
-
20
Word2Vec
Hierarchical Probabilistic
-
21
Word2Vec
-
22
...
N-gram K-meansAll Documents
BOWper Document
Word2vec
N
-
23
()
Average support vector regression
Support vector regression
+
= . .
-
24
()
-
25
?
-
26
multi-task learning
()task
Task 1 - feature
Task 2 - feature
Task 8 - feature
.
.
.
Kernel
Multi-task learning
-
27
= . .
?
-
28
- . . .
-
29
-
30
-
31
(Taiwan Triage and Acuity Scale, TTAS)
-
32
(=)
(~200)
(=)
(=)
(=)
-
33
:
()
SpeakerDiarization
-
Raw audio-videorecording
S1
S2
Sk
. . . MFCCPitch
Intensity
1 : [1,1]
2 : [1, 2]
: [1,]
34
:
:
S1
-
35
:
Support vector classification
Support vector classification
Fusion
-
36
:
72.3%
51.6%
gold standard
-
37
(: 0-3, : 4-6, : 7-10)
: :
: :
: :
Poker face Talk with smiling
Trembling voice
-
38
database
Before After
:
:
:
-
39
Before After
: :
: :
-
40
Pilot work ()
- . . .
-
41
( ~ 2-5s)
Global label ()
3-5 minutes
-
42
Thin-slice
Naumann et al. : Personality
Ovies et al. : Affect style
Oltmanns et al. : Personality disorders
-
:
Motion Capture(Avatar)
43
The USC CreativeIT database
-
44
:
:
: 45: 90
-
45
(multimodal)
(density-weighted)
(mutualinformation)
thin-slice
-
46
Activation 0.384 0.722
Dominance 0.675 0.834
Valence 0.571 0.822
(Global)
(Spearman )
-
47
91%
9%
Act.(10% data remain)
Including
Reduced 98%
2%
Dom.(70% data remain)
Including
Reduced
95%
5%
Val.(20% data remain)
Including
Reduced
thin-slice?
-
48
-
49
- :
1. 2.
(10)
-
Activation:
4.4 ()3.8 ()4.6 ()
0
1
1 3 5 7 9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
55
57
59
61
63
65
67
69
71
73
75
77
79
81
83
85
87
89
91
93
95
97
99
TIME SEGMENTS
Emotion-Rich behaviors
1
-
0
1
1 3 5 7 9
11
13
15
17
19
21
23
25
27
29
31
33
35
37
39
41
43
45
47
49
51
53
55
57
59
61
63
65
67
69
71
73
75
77
79
81
83
85
87
89
91
93
95
97
99
TIME SEGMENTS
Emotion-Rich behaviors
2
Valence: 4.0 ()3.7 (4.3 ()
-
52
Assumption: Gold Standard
-
53
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Act. Val.
Agreement
Entire Slice
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Act. Val.
Correlation
Entire Slice
thin slice
thin slice
-
54
-
55
-
56
()
Pattern
Contextualize
-
57
Data
evaluation
Always look for insights
-
58
-
59
ASD ADOS
Couple Therapy
Affective Computing
Oral Evaluation
Stroke Prediction
BiiC: BSP
fMRI Analysis
Pain Scale
-
60
-
61
:
application domain
-
62
Challenging the status quoMaking a positive impact
-
63
BiiC lab @ NTHU EEhttp://biic.ee.nthu.edu.tw