TKPERM: Cross-platform Permission Knowledge Transfer to Detect - - PowerPoint PPT Presentation

▶
tkperm cross platform permission knowledge transfer to
SMART_READER_LITE
LIVE PREVIEW

TKPERM: Cross-platform Permission Knowledge Transfer to Detect - - PowerPoint PPT Presentation

Jan 11, 2024 22 likes •845 views

TKPERM: Cross-platform Permission Knowledge Transfer to Detect Overprivileged Third-party Applications Faysal Hossain Shezan, Kaiming Cheng, Zhen Zhang, Yinzhi Cao, Yuan Tian Permission-based Access Control Android Chrome IFTTT 2

slide-1
SLIDE 1

TKPERM: Cross-platform Permission Knowledge Transfer to Detect Overprivileged Third-party Applications

Faysal Hossain Shezan, Kaiming Cheng, Zhen Zhang, Yinzhi Cao, Yuan Tian

slide-2
SLIDE 2

Permission-based Access Control

2

Android Chrome IFTTT

slide-3
SLIDE 3

Permission Correlation with Description

UBER- https://play.google.com/store/apps/details?id=com.ubercab&hl=en_US

3

Android App

slide-4
SLIDE 4

Permission Correlation with Description

UBER- https://play.google.com/store/apps/details?id=com.ubercab&hl=en_US

4

Location Permission Android App Requested Permission

slide-5
SLIDE 5

Permission Correlation with Description

UBER- https://play.google.com/store/apps/details?id=com.ubercab&hl=en_US

5

The app uses your location so your driver knows where to pick you up. Location Permission Android App Requested Permission Uber Description

slide-6
SLIDE 6

Permission Correlation with Description

UBER- https://play.google.com/store/apps/details?id=com.ubercab&hl=en_US

6

The app uses your location so your driver knows where to pick you up. Location Permission Android App Requested Permission Uber Description Consistent

slide-7
SLIDE 7

What is Overprivileged?

GamingHub- https://chrome.google.com/webstore/detail/gaminghub/eafoaklfmpnpdecnhhaailihkdbhkgin

7

GamingHub (Chrome Extension)

slide-8
SLIDE 8

What is Overprivileged?

GamingHub- https://chrome.google.com/webstore/detail/gaminghub/eafoaklfmpnpdecnhhaailihkdbhkgin

8

GamingHub (Chrome Extension) Location Permission Requested Permission

slide-9
SLIDE 9

What is Overprivileged?

GamingHub- https://chrome.google.com/webstore/detail/gaminghub/eafoaklfmpnpdecnhhaailihkdbhkgin

9

GamingHub (Chrome Extension) Location Permission Requested Permission GamingHub Description

Primary Features:

  • 1. Quick & Easy Access to popular

web games

  • 2. Minimalist & Elegant Design
  • 3. Hand Picked High Quality

Wallpapers that change according to mood

  • 4. New & Exciting ways for

accessing Online Content

  • 5. Let us know what you'd like,

more to come soon!

slide-10
SLIDE 10

What is Overprivileged?

GamingHub- https://chrome.google.com/webstore/detail/gaminghub/eafoaklfmpnpdecnhhaailihkdbhkgin

10

GamingHub (Chrome Extension) Location Permission No Explanation for the Usage of Location Permission Requested Permission GamingHub Description No Match

slide-11
SLIDE 11

Challenges

11

Taken from: https://iot-analytics.com/iot-platform-companies-landscape-2020/

slide-12
SLIDE 12

Challenges

Extensive data labeling and parameter tuning on new platforms Some platforms have limited data

12

Taken from: https://iot-analytics.com/iot-platform-companies-landscape-2020/

slide-13
SLIDE 13

Key Insights

Permission Knowledge Chrome App Android App

13

Location

slide-14
SLIDE 14

Solution- Transfer Learning

14

slide-15
SLIDE 15

Goal

General framework to detect unexpected permissions

15

slide-16
SLIDE 16

Research Questions

  • 1. What knowledge to transfer? (e.g., what original domain

should we select, what permissions in Android should we use)?

  • 2. How to minimize the amount of labeled data needed?

16

slide-17
SLIDE 17

System Overview of TKPERM

17

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform

slide-18
SLIDE 18

System Overview of TKPERM

18

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection 1

slide-19
SLIDE 19

System Overview of TKPERM

19

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection Source Model Training 1 2

slide-20
SLIDE 20

System Overview of TKPERM

20

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection Source Model Training Source Model 1 2 3

slide-21
SLIDE 21

System Overview of TKPERM

21

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Source Model 1 2 3

slide-22
SLIDE 22

System Overview of TKPERM

22

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Source Model 1 2 3 4 5

+

slide-23
SLIDE 23

System Overview of TKPERM

23

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Target Model Training Source Model 1 2 3 4 5 6 7

+ +

slide-24
SLIDE 24

System Overview of TKPERM

24

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Target Model Training Source Model Target Model 1 2 3 4 5 6 7 8

+ +

slide-25
SLIDE 25

System Overview of TKPERM

25

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Target Model Training Source Model Target Model 1 2 3 4 5 6 7 8

+ +

slide-26
SLIDE 26

Domain Selection

26

Compute and aggregate source domain(s) performs

Greedy Selection Approach

Research Question: What knowledge to transfer?

slide-27
SLIDE 27

Domain Selection

27

Compute and aggregate source domain(s) performs Remove source domain(s) which work worst

Greedy Selection Approach

Research Question: What knowledge to transfer?

slide-28
SLIDE 28

Domain Selection

28

Compute and aggregate source domain(s) performs Remove source domain(s) which work worst Find the best combination of the source domain(s)

Greedy Selection Approach

Research Question: What knowledge to transfer?

slide-29
SLIDE 29

Domain Selection

29

Compute and aggregate source domain(s) performs Remove source domain(s) which work worst Find the best combination of the source domain(s)

Greedy Selection Approach

Research Question: What knowledge to transfer?

slide-30
SLIDE 30

Data Selection

30

Use source model to rank the unlabeled document

Research Question: How to minimize the amount of labeled data needed?

slide-31
SLIDE 31

Data Selection

31

Use source model to rank the unlabeled document Pick the top 20 documents from the target domain

Research Question: How to minimize the amount of labeled data needed?

slide-32
SLIDE 32

Data Selection

32

Use source model to rank the unlabeled document Pick the top 20 documents from the target domain Ask human annotator to label data

Research Question: How to minimize the amount of labeled data needed?

slide-33
SLIDE 33

Data Selection

33

Use source model to rank the unlabeled document Pick the top 20 documents from the target domain Ask human annotator to label data

Research Question: How to minimize the amount of labeled data needed?

slide-34
SLIDE 34

Dataset

34

Available at: https://drive.google.com/drive/u/1/folders/1Yfnz-ZpBpL8lftYIdM6JtH-QKE88NcSX

292 Sentences 666 Sentences 4,705 Sentences SmartThings Chrome IFTTT 36,193 Sentences Android

slide-35
SLIDE 35

Dataset

35

AUTOCOG

AutoCog: Measuring the Description-to-permission Fidelity in Android Applications, Qu et al. (CCS 2014)

292 Sentences 666 Sentences 4,705 Sentences SmartThings Chrome IFTTT 36,193 Sentences Android

slide-36
SLIDE 36

Evaluation

Question 1. What is the end-to-end performance of TKPERM? Question 2. What is the performance of each component in TKPERM? Question 3. What is the computation overhead of TKPERM?

36

slide-37
SLIDE 37

Evaluation

Question 1. What is the end-to-end performance of TKPERM? Question 2. What is the performance of each component in TKPERM? Question 3. What is the computation overhead of TKPERM?

Effectiveness

37

slide-38
SLIDE 38

Evaluation (Effectiveness)

Source Domain Selection: H-divergence v/s Greedy Selection in IFTTT Platform

38

Target Domain Source Selection Source Domain(s) F1 Evernote H-Divergence Read Calendar 75.86% Greedy Selection Coarse Location + Fine Location + Camera 83.13% BMW Lab H-Divergence Read Contact 92.30% Greedy Selection Send SMS + Record Audio 95.24% Facebook H-Divergence Read Calendar 76.09% Greedy Selection Camera 88.09% Google Calendar H-Divergence Read Calendar 91.30% Greedy Selection Read Calendar + Coarse Location 92.30% Google Contact H-Divergence Read Contacts 99.20% Greedy Selection Read Contacts 99.20%

slide-39
SLIDE 39

Evaluation (Effectiveness)

Source Domain Selection: H-divergence v/s Greedy Selection in IFTTT Platform

39

Target Domain Source Selection Source Domain(s) F1 Evernote H-Divergence Read Calendar 75.86% Greedy Selection Coarse Location + Fine Location + Camera 83.13% BMW Lab H-Divergence Read Contact 92.30% Greedy Selection Send SMS + Record Audio 95.24% Facebook H-Divergence Read Calendar 76.09% Greedy Selection Camera 88.09% Google Calendar H-Divergence Read Calendar 91.30% Greedy Selection Read Calendar + Coarse Location 92.30% Google Contact H-Divergence Read Contacts 99.20% Greedy Selection Read Contacts 99.20%

slide-40
SLIDE 40

Evaluation (Effectiveness)

Data Selection: Comparison of With & Without Data Selection

40

Platform Performance Configuration No Transfer Without Data Selection With Data Selection IFTTT F1 Score 84.25% 91.08% 91.83% Improvement

  • 6.83%

7.58% Chrome F1 Score 70.60% 84.36% 89.13% Improvement

  • 13.76%

18.53% SmartThings F1 Score 72.80% 84.65% 89.1% Improvement

  • 11.85%

16.3%

slide-41
SLIDE 41

Evaluation (Effectiveness)

Data Selection: Comparison of With & Without Data Selection

41

Platform Performance Configuration No Transfer Without Data Selection With Data Selection IFTTT F1 Score 84.25% 91.08% 91.83% Improvement

  • 6.83%

7.58% Chrome F1 Score 70.60% 84.36% 89.13% Improvement

  • 13.76%

18.53% SmartThings F1 Score 72.80% 84.65% 89.1% Improvement

  • 11.85%

16.3%

slide-42
SLIDE 42

Platform Target Domain Source Domain Transfer No Transfer Improvement IFTTT Evernote Coarse Location + Fine Location + Camera 83.13% 79.78% 3.35% BMW Lab Send SMS + Record Audio 95.24% 85.71% 9.53% Facebook Camera 88.09% 75.00% 13.09% Google Calendar Read Calendar + Coarse Location 94.30% 83.54% 10.76% Google Contact Read Contact 98.41% 97.22% 1.19% Chrome Geolocation Fine Location + Coarse Location + Read Contact 88.29% 62.50% 25.79% Proxy Send SMS + Fine Location 93.78% 89.69% 4.09% Content Settings Fine Location + Read Contact 85.31% 59.61% 25.70% SmartThings Lock Write Setting 85.71% 75.00% 10.71% Motion Sensor Read Contact 87.10% 53.33% 33.77% Switch Send SMS + Read Calendar 94.39% 90.09% 4.30%

Evaluation (Effectiveness)

TKPERM Performance Analysis (Metric: F1 Score)

42

slide-43
SLIDE 43

Platform Target Domain Source Domain Transfer No Transfer Improvement IFTTT Evernote Coarse Location + Fine Location + Camera 83.13% 79.78% 3.35% BMW Lab Send SMS + Record Audio 95.24% 85.71% 9.53% Facebook Camera 88.09% 75.00% 13.09% Google Calendar Read Calendar + Coarse Location 94.30% 83.54% 10.76% Google Contact Read Contact 98.41% 97.22% 1.19% Chrome Geolocation Fine Location + Coarse Location + Read Contact 88.29% 62.50% 25.79% Proxy Send SMS + Fine Location 93.78% 89.69% 4.09% Content Settings Fine Location + Read Contact 85.31% 59.61% 25.70% SmartThings Lock Write Setting 85.71% 75.00% 10.71% Motion Sensor Read Contact 87.10% 53.33% 33.77% Switch Send SMS + Read Calendar 94.39% 90.09% 4.30%

Evaluation (Effectiveness)

TKPERM Performance Analysis (Metric: F1 Score)

43

slide-44
SLIDE 44

Platform Target Domain Source Domain Transfer No Transfer Improvement IFTTT Evernote Coarse Location + Fine Location + Camera 83.13% 79.78% 3.35% BMW Lab Send SMS + Record Audio 95.24% 85.71% 9.53% Facebook Camera 88.09% 75.00% 13.09% Google Calendar Read Calendar + Coarse Location 94.30% 83.54% 10.76% Google Contact Read Contact 98.41% 97.22% 1.19% Chrome Geolocation Fine Location + Coarse Location + Read Contact 88.29% 62.50% 25.79% Proxy Send SMS + Fine Location 93.78% 89.69% 4.09% Content Settings Fine Location + Read Contact 85.31% 59.61% 25.70% SmartThings Lock Write Setting 85.71% 75.00% 10.71% Motion Sensor Read Contact 87.10% 53.33% 33.77% Switch Send SMS + Read Calendar 94.39% 90.09% 4.30%

Evaluation (Effectiveness)

TKPERM Performance Analysis (Metric: F1 Score)

44

slide-45
SLIDE 45

Platform Target Domain Source Domain Transfer No Transfer Improvement IFTTT Evernote Coarse Location + Fine Location + Camera 83.13% 79.78% 3.35% BMW Lab Send SMS + Record Audio 95.24% 85.71% 9.53% Facebook Camera 88.09% 75.00% 13.09% Google Calendar Read Calendar + Coarse Location 94.30% 83.54% 10.76% Google Contact Read Contact 98.41% 97.22% 1.19% Chrome Geolocation Fine Location + Coarse Location + Read Contact 88.29% 62.50% 25.79% Proxy Send SMS + Fine Location 93.78% 89.69% 4.09% Content Settings Fine Location + Read Contact 85.31% 59.61% 25.70% SmartThings Lock Write Setting 85.71% 75.00% 10.71% Motion Sensor Read Contact 87.10% 53.33% 33.77% Switch Send SMS + Read Calendar 94.39% 90.09% 4.30%

Evaluation (Effectiveness)

TKPERM Performance Analysis (Metric: F1 Score)

45

12.93% improvement compared to No Transfer

slide-46
SLIDE 46

Evaluation

Question 1. What is the end-to-end performance of TKPERM? Question 2. What is the performance of each component in TKPERM? Question 3. What is the computation overhead of TKPERM?

Effectiveness

46

slide-47
SLIDE 47

Evaluation

Question 1. What is the end-to-end performance of TKPERM? Question 2. What is the performance of each component in TKPERM? Question 3. What is the computation overhead of TKPERM?

Effectiveness Scalability

47

slide-48
SLIDE 48

Evaluation (Scalability)

Computation Overhead (Run in Amazon Elastic Compute Cloud (EC2), NVIDIA Tesla V100)

48

Platform Target Domain Time (hh:mm:ss) IFTTT Evernote 33:27:03 BMW Lab 14:08:40 Facebook 22:57:20 Google Calendar 15:15:18 Google Contact 18:40:17 Chrome Geolocation 07:37:28 Proxy 06:54:01 Content Settings 09:42:45 SmartThings Lock 03:47:59 Motion Sensor 04:09:44 Switch 14:11:08

slide-49
SLIDE 49

Evaluation (Scalability)

Computation Overhead (Run in Amazon Elastic Compute Cloud (EC2), NVIDIA Tesla V100)

49

Platform Target Domain Time (hh:mm:ss) IFTTT Evernote 33:27:03 BMW Lab 14:08:40 Facebook 22:57:20 Google Calendar 15:15:18 Google Contact 18:40:17 Chrome Geolocation 07:37:28 Proxy 06:54:01 Content Settings 09:42:45 SmartThings Lock 03:47:59 Motion Sensor 04:09:44 Switch 14:11:08

slide-50
SLIDE 50

Measurement Result

50

114 Chrome Apps (35.73%)

slide-51
SLIDE 51

Measurement Result

51

135 IFTTT Apps (28.36%) 114 Chrome Apps (35.73%)

slide-52
SLIDE 52

Measurement Result

52

80 SmartThings Apps (32.9%) 135 IFTTT Apps (28.36%) 114 Chrome Apps (35.73%)

slide-53
SLIDE 53

Measurement Result

53

80 SmartThings Apps (32.9%) 135 IFTTT Apps (28.36%) 114 Chrome Apps (35.73%) 329 Overprivileged Apps (32.33%)

slide-54
SLIDE 54

Conclusion

54

General framework to detect Overprivileged applications in new platforms

  • 1. General Framework
slide-55
SLIDE 55
  • 1. General Framework

Conclusion

TKPERM works well (90.02% F1 score on avg.)

Posted in- https://www.reddit.com/r/GearVR/comments/5ga1na/just_got_the_vr_why_do_some_apps_ask_for_so_many/

55

slide-56
SLIDE 56
  • 1. General Framework

Conclusion

TKPERM works well (90.02% F1 score on avg.)

Posted in- https://www.reddit.com/r/GearVR/comments/5ga1na/just_got_the_vr_why_do_some_apps_ask_for_so_many/

How to identify overprivileged application in VR (new platform)?

56

slide-57
SLIDE 57
  • 1. General Framework

Conclusion

Posted in- https://www.reddit.com/r/GearVR/comments/5ga1na/just_got_the_vr_why_do_some_apps_ask_for_so_many/

TKPERM!!!

57

slide-58
SLIDE 58

Conclusion

TKPERM works well (90.02% F1 score on avg.)

58

  • 1. General Framework
  • 2. Result
slide-59
SLIDE 59

Conclusion

59

  • 1. General Framework
  • 2. Result
  • 3. Public Dataset

Dataset Available at: https://drive.google.com/drive/u/1/folders/1Yfnz-ZpBpL8lftYIdM6JtH-QKE88NcSX

slide-60
SLIDE 60

Thank You!

Contact: Faysal Hossain Shezan (Email-fs5ve@virginia.edu)

60

slide-61
SLIDE 61

Conclusion

Email: fs5ve@virginia.edu

61

Dataset Available at: https://drive.google.com/drive/u/1/folders/1Yfnz-ZpBpL8lftYIdM6JtH-QKE88NcSX

  • 1. General Framework
  • 2. Result
  • 3. Public Dataset

Dataset

slide-62
SLIDE 62

Backup Slides

62

slide-63
SLIDE 63

82

slide-64
SLIDE 64

83

slide-65
SLIDE 65

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection 1

System Overview (Domain Selection)

91

slide-66
SLIDE 66

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Source Model 1 2 3 4 5

System Overview (Data Selection)

Ranking: Selection:

96

slide-67
SLIDE 67

Permission-based Access Control

108

Android Chrome IFTTT

slide-68
SLIDE 68

Goal

  • G1. Semantic Knowledge

Android Chrome IFTTT SmartThings

109

slide-69
SLIDE 69

Goal

  • G1. Semantic Knowledge
  • G2. Permission Knowledge

Android Chrome IFTTT SmartThings

110

slide-70
SLIDE 70

System Overview

111

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform

slide-71
SLIDE 71

System Overview

112

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform

Challenge: What knowledge to transfer?

slide-72
SLIDE 72

Domain Selection

113

Aggregate source domain(s) which performs best Remove source domain(s) which work worst Find the best combination of the source domain(s)

Greedy Selection Approach

slide-73
SLIDE 73

System Overview

114

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection 1

slide-74
SLIDE 74

System Overview

115

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection Source Model Training 1 2

slide-75
SLIDE 75

System Overview

116

Read Contacts Access Coarse Location Access Fine Location Camera ………. Source Platform Domain Selection Source Model Training Source Model 1 2 3

slide-76
SLIDE 76

System Overview

117

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Source Model 1 2 3

slide-77
SLIDE 77

System Overview

118

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Source Model 1 2 3

Challenge: How to minimize the amount of labeled data needed?

slide-78
SLIDE 78

Data Selection

119

Use source model to rank the document Rank unlabeled documents from the target domain Pick the top 20 documents from a target domain Ask human annotator to label data

slide-79
SLIDE 79

System Overview

120

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Source Model 1 2 3 4 5

+

slide-80
SLIDE 80

System Overview

121

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Target Model Training Source Model 1 2 3 4 5 6 7

+ +

slide-81
SLIDE 81

System Overview

122

Read Contacts Access Coarse Location Access Fine Location Camera ………. Chrome Geolocation Chrome Proxy Chrome Content Settings SmartThings Switch ………. Source Platform Target Platforms Domain Selection Source Model Training Data Selection Target Model Training Source Model Target Model 1 2 3 4 5 6 7 8

+ +

slide-82
SLIDE 82

Conclusion

  • IFTTT – 135 apps
  • Chrome Extension – 114 apps
  • SmartThings – 80 apps

127