Towards Next Generation Secure DDDAS/Infosymbiotics Systems
ICCS 2015, Reykjavik, Iceland June 2015
Towards Next Generation Secure DDDAS/Infosymbiotics Systems Li Xiong and Vaidy Sunderam Students: Layla Pournajaf, Daniel Garcia-Ulloa, Xiaofeng Xu
Dept. of Math and Computer Science Emory University
AFOSR DDDAS FA9550-12-1-0240
DDDAS as a Unifying Paradigm • Ability to dynamically integrate generated data into an application; feedback loop to steer measurement • • • •
Acquisition – measurements, streams, databases Assimilation – preprocessing, aggregation, fusion Analytics – simulations, decisions, knowledge discovery Action – incorporate new results, feedback to above
• Platforms & Domains • • • • •
Internet of Things (IoT), Smart(er) Systems Physical, chemical, biological, engineering, weather Medical, health, transport, infrastructure, military, disaster Trends: InfoSymbiotics – Big data and Big computing Evolution: ubiquitous sensing/informatics/multimodal
From the Sensor-Scale to the Exa-Scale • Hierarchical DDDAS • Devices • • • •
Embedded devices Sensors UAV/UGV Participants
• Regional/Central • HPC Clusters • Exascale machines • Data/knowledge bases
• Networking
Multilevel DDDAS Systems • End-to-end data/compute/control flow & interaction
*Original figure due to Dr. Frederica Darema
Next Generation DDDAS/InfoSymbiotics Systems • Participant/data privacy • Identity, location and data are all sensitive
• Uncertainty • Measurements/observations subject to error • At exascale, intermittent failures are inevitable • Cloaking/obfuscation for privacy
• Handle privacy & uncertainty within unified rubric • Aggregation, fusion and summarization • Transformations in the presence of uncertainty
• Secure high-performance multiparty computation • At each DDDAS level, perform local computations and analytics, cooperatively with mutually untrusted peers
Foundational Work • Privacy Preserving Data Collection with Feedback Control • Privacy Preserving Data Aggregation with Feedback Control • Secure Data Collection and Aggregation Privacy Preserving Feedback Control
Cloaking
Aggregation
Prediction
Collection
Perturbation
Correction
Privacy Preserving Data Collection
Data Contributors
Sensitive Data Streams
Privacy Preserving Data Aggregation
Trusted Aggregator
Aggregated Data streams
Data Modeling
Application
Next Generation DDDAS
}
• Privacy-preserving, secure acquisition High-performance • Fusion/aggregation of uncertain data secure distr. comp. • Prediction/correction/application steering + feedback loop
Privacy Preserving Participant Management • Feedback-controlled assignment of cloaked mobile participants to targets
Task management feedback
Measurement feedback
Input/steering data
• Challenges: maximize coverage, minimize cost; handle mobile participants/targets
DDDAS Feedback-driven Tasking
a) Exact Trajectories Predictive/Corrective scheme augmented with mobility model
b) Uncertain Trajectories
Model: Xt ∼ p(Xt | Xt−1) Zt ∼ p(Zt | Xt) Meas: Z1:t = Z1, . . . , Zt Pred:
p(Xt | Z1:t−1) = Σ p(Xt | Xt−1) p(Xt−1 | Z1:t−1)
Update: p(Xt | Z1:t) = p(Yt | Xt) p(Xt | Z1:t−1) Σ p(Yt | Xt) p(Xt | Z1:t−1)
Data Assimilation under Uncertainty • Objective: Aggregation/fusion of unreliable observations for analytics/decision-making • Spatio-temporal crowdsensing example: • M participants (unreliably) report about • N events at one or more of R consecutive times • Observations ∈ S = {s1, s2, … sv} or ∅ (missing)
• Determine “state label” at location lj at time tk
Truth Inference Approach • Hidden Markov Model using iterative approach to determine transition probabilities
• Algorithm summary • Initial guess history + heuristics • Seek max posterior probability • Semi- and un-supervised learning
• Challenges: methods for other aggregation/ fusion/assimilation functions with uncertain data
High-performance Distributed SMC • Secure Multi-Party Computation • Guarantees that computation does not reveal private input
• Possible approaches • Shamir’s secret sharing scheme • Perturbation based • Homomorphic encryption schemes
• Efficiency (secure sum)
12
DDDAS Software Toolkit • Scalable and stateless distributed computing • • • •
Small footprint for sensors and field devices Low latency, low power communications Adopt models/features from FreshBreeze/ROS/HELib Deployable at field regional levels, interfaces to traditional supercomputer simulations
• Algorithm libraries for SMC, distributed computation • Building block modules (multiplication, division, matrix inversion) • Higher level functions (distributed Kalman filter, statistical summarization, global optimization functions)
• Challenge: robust uncertainty-resilient implementations adaptively balancing utility (accuracy) and efficiency 13
Summary • Next generation DDDAS/Infosymbiotics systems • Ever expanding platforms – Internet of Things, Smart Systems • Unified systems/software model for numerous applications
• Requirements and expectations • Privacy and security – of participants, data, computation • Uncertainty – resilience to errors, faults, obfuscation, (mis)trust • Autonomous local and hierarchical analytics, decision makeing
• The PREDICT project • Feedback driven dynamic management of sensor-participant systems with privacy protection • Trust-aware data synthesis, aggregation and validation • Secure high-performance distributed computing software
Thank you • Acknowledgements • AFOSR DDDAS FA9550-12-1-0240
• Project team • Investigators: Li Xiong, Vaidy Sunderam • Students: Liyue Fan, Slawek Goryczka, Layla Pournjaf, Daniel Garcia-Ulloa, Xiaofeng Xu
• Project URL • http://www.mathcs.emory.edu/predict/
AFOSR DDDAS FA9550-12-1-0240
Towards Next Generation Secure DDDAS/Infosymbiotics Systems Li Xiong and Vaidy Sunderam Students: Layla Pournajaf, Daniel Garcia-Ulloa, Xiaofeng Xu
Dept. of Math and Computer Science Emory University
AFOSR DDDAS FA9550-12-1-0240
DDDAS as a Unifying Paradigm • Ability to dynamically integrate generated data into an application; feedback loop to steer measurement • • • •
Acquisition – measurements, streams, databases Assimilation – preprocessing, aggregation, fusion Analytics – simulations, decisions, knowledge discovery Action – incorporate new results, feedback to above
• Platforms & Domains • • • • •
Internet of Things (IoT), Smart(er) Systems Physical, chemical, biological, engineering, weather Medical, health, transport, infrastructure, military, disaster Trends: InfoSymbiotics – Big data and Big computing Evolution: ubiquitous sensing/informatics/multimodal
From the Sensor-Scale to the Exa-Scale • Hierarchical DDDAS • Devices • • • •
Embedded devices Sensors UAV/UGV Participants
• Regional/Central • HPC Clusters • Exascale machines • Data/knowledge bases
• Networking
Multilevel DDDAS Systems • End-to-end data/compute/control flow & interaction
*Original figure due to Dr. Frederica Darema
Next Generation DDDAS/InfoSymbiotics Systems • Participant/data privacy • Identity, location and data are all sensitive
• Uncertainty • Measurements/observations subject to error • At exascale, intermittent failures are inevitable • Cloaking/obfuscation for privacy
• Handle privacy & uncertainty within unified rubric • Aggregation, fusion and summarization • Transformations in the presence of uncertainty
• Secure high-performance multiparty computation • At each DDDAS level, perform local computations and analytics, cooperatively with mutually untrusted peers
Foundational Work • Privacy Preserving Data Collection with Feedback Control • Privacy Preserving Data Aggregation with Feedback Control • Secure Data Collection and Aggregation Privacy Preserving Feedback Control
Cloaking
Aggregation
Prediction
Collection
Perturbation
Correction
Privacy Preserving Data Collection
Data Contributors
Sensitive Data Streams
Privacy Preserving Data Aggregation
Trusted Aggregator
Aggregated Data streams
Data Modeling
Application
Next Generation DDDAS
}
• Privacy-preserving, secure acquisition High-performance • Fusion/aggregation of uncertain data secure distr. comp. • Prediction/correction/application steering + feedback loop
Privacy Preserving Participant Management • Feedback-controlled assignment of cloaked mobile participants to targets
Task management feedback
Measurement feedback
Input/steering data
• Challenges: maximize coverage, minimize cost; handle mobile participants/targets
DDDAS Feedback-driven Tasking
a) Exact Trajectories Predictive/Corrective scheme augmented with mobility model
b) Uncertain Trajectories
Model: Xt ∼ p(Xt | Xt−1) Zt ∼ p(Zt | Xt) Meas: Z1:t = Z1, . . . , Zt Pred:
p(Xt | Z1:t−1) = Σ p(Xt | Xt−1) p(Xt−1 | Z1:t−1)
Update: p(Xt | Z1:t) = p(Yt | Xt) p(Xt | Z1:t−1) Σ p(Yt | Xt) p(Xt | Z1:t−1)
Data Assimilation under Uncertainty • Objective: Aggregation/fusion of unreliable observations for analytics/decision-making • Spatio-temporal crowdsensing example: • M participants (unreliably) report about • N events at one or more of R consecutive times • Observations ∈ S = {s1, s2, … sv} or ∅ (missing)
• Determine “state label” at location lj at time tk
Truth Inference Approach • Hidden Markov Model using iterative approach to determine transition probabilities
• Algorithm summary • Initial guess history + heuristics • Seek max posterior probability • Semi- and un-supervised learning
• Challenges: methods for other aggregation/ fusion/assimilation functions with uncertain data
High-performance Distributed SMC • Secure Multi-Party Computation • Guarantees that computation does not reveal private input
• Possible approaches • Shamir’s secret sharing scheme • Perturbation based • Homomorphic encryption schemes
• Efficiency (secure sum)
12
DDDAS Software Toolkit • Scalable and stateless distributed computing • • • •
Small footprint for sensors and field devices Low latency, low power communications Adopt models/features from FreshBreeze/ROS/HELib Deployable at field regional levels, interfaces to traditional supercomputer simulations
• Algorithm libraries for SMC, distributed computation • Building block modules (multiplication, division, matrix inversion) • Higher level functions (distributed Kalman filter, statistical summarization, global optimization functions)
• Challenge: robust uncertainty-resilient implementations adaptively balancing utility (accuracy) and efficiency 13
Summary • Next generation DDDAS/Infosymbiotics systems • Ever expanding platforms – Internet of Things, Smart Systems • Unified systems/software model for numerous applications
• Requirements and expectations • Privacy and security – of participants, data, computation • Uncertainty – resilience to errors, faults, obfuscation, (mis)trust • Autonomous local and hierarchical analytics, decision makeing
• The PREDICT project • Feedback driven dynamic management of sensor-participant systems with privacy protection • Trust-aware data synthesis, aggregation and validation • Secure high-performance distributed computing software
Thank you • Acknowledgements • AFOSR DDDAS FA9550-12-1-0240
• Project team • Investigators: Li Xiong, Vaidy Sunderam • Students: Liyue Fan, Slawek Goryczka, Layla Pournjaf, Daniel Garcia-Ulloa, Xiaofeng Xu
• Project URL • http://www.mathcs.emory.edu/predict/
AFOSR DDDAS FA9550-12-1-0240
