C3P: Context-Aware Crowdsourced Cloud Privacy Privacy Enhancing Technologies Symposium, 2014 1 CloudSpaces
Files to Flowers Conversion 2
Files to Flowers Conversion 2
Files to Flowers Conversion 2
Files to Flowers Conversion 2
Files to Flowers Conversion 2
60% increase in corporate data shared to the cloud in 2015 3 Source: Elastica’s Q2 2015 Shadow Data Report
60% increase in corporate data shared to the cloud in 2015 20% of files shared to the cloud contain protected data 3 Source: Elastica’s Q2 2015 Shadow Data Report
60% increase in corporate data shared to the cloud in 2015 20% of files shared to the cloud contain protected data 60% 30% of sensitive files contain PII …contain health info 3 Source: Elastica’s Q2 2015 Shadow Data Report
60% increase in corporate data shared to the cloud in 2015 20% of files shared to the cloud contain protected data 60% 30% of sensitive files contain PII …contain health info Emergence of “Shadow IT” 3 Source: Elastica’s Q2 2015 Shadow Data Report
Anti-Snooping Tools for the Cloud Your files are always encrypted You are fully You cannot use before protected. cloud services. uploading. Examples: 4
What if Antivirus Software was Similar? Your files are You cannot You are fully always run protected. quarantined. software. 5
Obstacles Privacy vs. Services dilemma
Obstacles Privacy vs. Services dilemma Context-dependence of privacy
Obstacles Privacy vs. Services dilemma Context-dependence of privacy I dedicate the rest of Manual effort and expertise my life for sorting out sensitive from non-sensitive for assessing data sensitivity files on my HD 6
What is needed? Ensure serviceable protection instead of brute encryption.
What is needed? Ensure serviceable protection instead of brute encryption. Account for the metadata, sharing environment, and data content.
What is needed? Ensure serviceable protection instead of brute encryption. Account for the metadata, sharing environment, and data content. Automatically estimate the I dedicate the rest of my life for sorting out sensitive from non-sensitive sensitivity of shared data. files on my HD 7
Introducing C3P Various levels of information hiding 8
Introducing C3P Various levels of information hiding Define data in terms of context 8
Introducing C3P Various levels of information hiding Define data in terms of context Private crowdsourcing mechanism for gathering people privacy policies I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD 8
Introducing C3P Various levels of information hiding Define data in terms of context Private crowdsourcing mechanism for gathering people privacy policies I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD Psychologically grounded approach for estimating sensitivity 8
Fine-Grained Policies 9
Defining Data through Context Content Metadata Environment 10
Defining Data through Context Content Metadata Environment 10
Defining Data through Context Content Metadata Environment 10
Context V ocabulary Home Document Location Office Data Software Financial Media Topic Educational 11
Privacy Preserving Crowdsourcing I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD Financial Me Stranger Faces Home Friend Business Me Colleague Business Me Colleague Financial Me Stranger Faces Home Friend User 1 User 2 User 3 12
Privacy Preserving Crowdsourcing I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD Financial Me Stranger Faces Home Friend Business Me Colleague Context Sharing Operation Faces Home Friend Business Me Colleague Financial Me Stranger Faces Home Friend User 1 User 2 User 3 12
Privacy Preserving Crowdsourcing I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD Financial Me Stranger Faces Home Friend Business Me Colleague Family Context Sharing Operation Colleague Faces Home Friend Work Sea Business Me Colleague Financial Me Stranger Faces Home Friend User 1 User 2 User 3 12
Privacy Preserving Crowdsourcing I dedicate the rest of my life for sorting out sensitive from non-sensitive files on my HD Financial Me Stranger Faces Home Friend Business Me Colleague Family Context Context Sharing Operation Colleague Faces Home Friend Faces Home Friend K-anonymity Work Forward-Anonymity Sea Business Me Colleague Financial Me Stranger Faces Home Friend User 1 User 2 User 3 12
Sensitivity Estimation using Item Response Theory Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High Sensitivity 75% Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High Sensitivity 75% Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High High Sensitivity Privacy Attitude 75% 75% Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High High Sensitivity Privacy Attitude 75% 75% Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High High Sensitivity Privacy Attitude 75% 75% Faces Home Friend 13
Sensitivity Estimation using Item Response Theory High High Sensitivity Privacy Attitude 75% 75% Faces Home Friend Faces Home Friend Faces Home Friend Group Invariance 13
Sensitivity Estimation using Item Response Theory High High Sensitivity Privacy Attitude 75% 75% Faces Home Friend Faces Home Friend Faces Home Friend Item Group Invariance Invariance 13
Server Connecting the Dots ? Client 14
Server Connecting the Dots ? Client 14
Server Connecting the Dots ? Context Extraction Financial Me Stranger Client 14
Server Connecting the Dots Sensitivity Request ? Context Extraction Financial Me Stranger Client 14
Server Connecting the Dots ? Sensitivity Reply Financial Me Stranger Client 14
Server Connecting the Dots ? Sensitivity Reply Financial Me Stranger Policy Decision Client 14
Server Connecting the Dots ? Data Sharing Financial Me Stranger Policy Decision Client 14
Server Connecting the Dots ? Financial Me Stranger Crowdsourcing Client 14
Server Connecting the Dots ? Sensitivity Computation Financial Me Stranger Crowdsourcing Client 14
C3P Evaluation 15
IRT Models Fit Privacy-Aware Cloud Sharing? 96 81 16
IRT Models Fit Privacy-Aware Cloud Sharing? 96 81 • Ex: With which privacy level would you share a project presentation with a friend? 16
IRT Models Fit Privacy-Aware Cloud Sharing? Dichotomous case A dot represents a 96 Sensitivity context 81 • Ex: With which privacy level would you share a project presentation with a friend? Infit t-statistic • Standardized Infit Statistic: • (x-axis values should lie in [-2,2]) 16
IRT Models Fit Privacy-Aware Cloud Sharing? Dichotomous case A dot represents a 96 Sensitivity context 81 • Ex: With which privacy level would you share a project presentation with a friend? Infit t-statistic • Standardized Infit Statistic: • (x-axis values should lie in [-2,2]) 16
IRT Models Fit Privacy-Aware Cloud Sharing? Dichotomous case A dot represents a 96 Sensitivity context 81 • Ex: With which privacy level would you share a project presentation with a friend? Infit t-statistic Polytomous case • Standardized Infit Statistic: • (x-axis values should lie in [-2,2]) Sensitivity Infit t-statistic 16
IRT Models Fit Privacy-Aware Cloud Sharing? Dichotomous case A dot represents a 96 Sensitivity context 81 • Ex: With which privacy level would you share a project presentation with a friend? Infit t-statistic Polytomous case • Standardized Infit Statistic: • (x-axis values should lie in [-2,2]) Sensitivity Infit t-statistic 16
IRT Models Fit Privacy-Aware Cloud Sharing? Dichotomous case A dot represents a 96 Sensitivity context 81 • Ex: With which privacy level would you share a project presentation with a friend? Infit t-statistic Polytomous case • Standardized Infit Statistic: • (x-axis values should lie in [-2,2]) Sensitivity Yes! Infit t-statistic 16
Temporal Cost of Crowdsourcing & Privacy • Synthetic Dataset: 3125 500 30000 Zipf context distribution av.: 1 Item/6 hours 17
Temporal Cost of Crowdsourcing & Privacy • Synthetic Dataset: 3125 500 30000 k Zipf context distribution av.: 1 Item/6 hours 17
Temporal Cost of Crowdsourcing & Privacy • Synthetic Dataset: 3125 500 30000 k Zipf context distribution av.: 1 Item/6 hours 17 Crowdsourcing cost : Hit rate (HR) from 0 to 90% in 10 days
Recommend
More recommend
