nus computing - privacy-preserving sensor cloudchangec/publications/2016... · 2017-04-12 ·...
TRANSCRIPT
![Page 1: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/1.jpg)
Privacy-Preserving Sensor Cloud
Hung Dang, Yun Long Chong, Francois Brun, Ee-Chien ChangSchool of Computing
National University of Singapore
![Page 2: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/2.jpg)
Motivation
➢ The ubiquity of time series/multimedia data.➢ Privacy concerns.➢ The needs of sharings and/or collaboration.
![Page 3: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/3.jpg)
Application Scenario*
Sensor Cloud:
➢ Sensors are spatially arranged.
*Our techniques is also applicable to other applications involve multidimensional data.
![Page 4: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/4.jpg)
Application Scenario*
Sensor Cloud:
➢ Sensors are spatially arranged.➢ Sensors continuously sense,
encrypt and stream samples to the cloud.
User Ownerq
S
kS
S
c1, c
2,
…, cn
Sensorc
i
setup
*Our techniques is also applicable to other applications involve multidimensional data.
![Page 5: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/5.jpg)
Application Scenario*
Sensor Cloud:
➢ Sensors are spatially arranged.➢ Sensors continuously sense,
encrypt and stream samples to the cloud.
➢ Samples are indexed by temporal and spatial meta-information.
User Ownerq
S
kS
S
c1, c
2,
…, cn
Sensorc
i
setup
*Our techniques is also applicable to other applications involve multidimensional data.
![Page 6: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/6.jpg)
Application Scenario*
Sensor Cloud:
➢ Sensors are spatially arranged.➢ Sensors continuously sense,
encrypt and stream samples to the cloud.
➢ Samples are indexed by temporal and spatial meta-information.
➢ Sharings is done in query-and-response fashion: a query specifies a desired set of samples, a response grants access to the desired set.
User Ownerq
S
kS
S
c1, c
2,
…, cn
Sensorc
i
setup
*Our techniques is also applicable to other applications involve multidimensional data.
![Page 7: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/7.jpg)
System model
User Ownerq
S
kS
S
c1, c
2,
…, cn
Sensorc
i
setup
![Page 8: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/8.jpg)
Query Types
➢ Q1 - d-dimensional range query➢ Samples' indices form a d-dimensional.
e.g.: all samples on street A on date X.
Q2 - Down-sampling query
Samples' incides form a down-sampled lattice.
e.g.: Y samples per each hours on street A on date X.
Q3 - General query
Samples' indices may or may not have any structure.
e.g.: random set of samples captured on date X.
![Page 9: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/9.jpg)
Query Types
➢ Q1 - d-dimensional range query➢ Samples' indices form a d-dimensional.
e.g.: all samples on street A on date X.
➢ Q2 - Down-sampling query➢ Samples' incides form a down-sampled lattice.
e.g.: Y samples per each hours on street A on date X.
Q3 - General query
Samples' indices may or may not have any structure.
e.g.: random set of samples captured on date X.
![Page 10: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/10.jpg)
Query Types
➢ Q1 - d-dimensional range query➢ Samples' indices form a d-dimensional.
e.g.: all samples on street A on date X.
➢ Q2 - Down-sampling query➢ Samples' incides form a down-sampled lattice.
e.g.: Y samples per each hours on street A on date X.
➢ Q3 - General query➢ Samples' indices may or may not have any
structure.
e.g.: random set of samples captured on date X.
![Page 11: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/11.jpg)
Problem Definition
➢ Security Requirements:➢ Confidentiality of the samples.➢ Collusion resistance.
➢ combining multiple aggregated keys could not derive more information than each aggregated key can individually derive
➢ Sensors are trusted and independent.
Efficiency Requirements:
Low computation load.
Low communication overhead.
Low storage overhead.
![Page 12: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/12.jpg)
Problem Definition
➢ Security Requirements:➢ Confidentiality of the samples.➢ Collusion resistance.
➢ combining multiple aggregated keys could not derive more information than each aggregated key can individually derive
➢ Sensors are trusted and independent.
➢ Efficiency Requirements:➢ Low computation load.➢ Low communication overhead.➢ Low storage overhead. User Owner
qS
kS
S
c1, c
2,
…, cn
Sensorc
i
setup
![Page 13: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/13.jpg)
Our Solution
➢ Each sample is encrypted individually using unique key to avoid collusion attack.
Leverage on KAC to ensure:
Aggregating any set of keys into one constant size key, attaining low communication overhead.
Low storage overhead by constant size ciphertexts.
Propose fast reconstruction techniques to reduce the computation load.
Achieving orders of magnitude speed-up over original KAC.
![Page 14: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/14.jpg)
Our Solution
➢ Each sample is encrypted individually using unique key to avoid collusion attack.
➢ Leverage on KAC to ensure:➢ Aggregating any set of keys into one constant size key,
attaining low communication overhead.➢ Low storage overhead by constant size ciphertexts.
Propose fast reconstruction techniques to reduce the computation load.
Achieving orders of magnitude speed-up over original KAC.
![Page 15: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/15.jpg)
Our Solution
➢ Each sample is encrypted individually using unique key to avoid collusion attack.
➢ Leverage on KAC to ensure:➢ Aggregating any set of keys into one constant size key,
attaining low communication overhead.➢ Low storage overhead by constant size ciphertexts.
➢ Propose fast reconstruction techniques to reduce the computation load.➢ Achieving orders of magnitude speed-up over original
KAC.
![Page 16: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/16.jpg)
KAC Reconstruction Review
➢ Reconstructing a ciphertext with index i ϵ S using an aggregated key kS requires:
where all can be drawn from public parametters and is system capacity.
➢ This incurs O(|S|2) group multiplications to reconstruct all samples in S.
![Page 17: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/17.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
where
![Page 18: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/18.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 19: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/19.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 20: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/20.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 21: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/21.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 22: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/22.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 23: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/23.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 24: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/24.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 25: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/25.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 26: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/26.jpg)
A Key Observation
The recurrence relation
➢ How many multiplications to evaluate X?
It is O(m), not O(m2)➢
where
![Page 27: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/27.jpg)
Fast Reconstruction for Q1
For Q1 with S = [1,m]:➢
➢ A special recurrence relation:
i.e. obtaining from with two multiplications.
=> In general, reconstructing samples in d-dimensional range query requires only O(d|S|) multiplications; i.e. linear time.
![Page 28: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/28.jpg)
Fast Reconstruction for Q1
For e.g., with S [1..5], system capacity n = 20:
![Page 29: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/29.jpg)
Fast Reconstruction for Q1
For e.g., with S [1..5], system capacity n = 20:
![Page 30: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/30.jpg)
Fast Reconstruction for Q1
For e.g., with S [1..5], system capacity n = 20:
![Page 31: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/31.jpg)
Fast Reconstruction for Q1
For e.g., with S [1..5], system capacity n = 20:
![Page 32: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/32.jpg)
Fast Reconstruction for Q2
Transform and Conquer strategy:➢ Transform the coordinate system such that indices
of the required samples correspond to integer coordinates.
➢ Apply the special recurrence relation as in Q1.
=> Also requires only O(d|S|) multiplications; i.e. linear time.
![Page 33: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/33.jpg)
Fast Reconstruction for Q3
➢ Samples' indices in Q3 may not have an special structure, to which the special recurrence could not apply.
➢ Problem transformation:➢ Let Pi be a multi-set comprising of all required to
compute , T the target collection comprising of all Pi.
➢ A computation plan to evaluate all is equivalent to that of constructing T.
![Page 34: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/34.jpg)
Fast Reconstruction for Q3
Minimum Spanning Tree based Strategy:➢ Define dist(i, j) = |Pi \Pj| + |Pj \Pi |
➢ A computation plan is determined by solving for the MST on a graph G = (V,E):
➢ G is complete.➢ V comprises of |T|+1 vertices: Vertex vi represent a multiset
Pi, and special vertex v represents empty multiset.
➢ An edge eij connecting vi and vj has weigh of dist(i,j). All edges orignating from v have weight of |T| - 2.
![Page 35: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/35.jpg)
Fast Reconstruction for Q3
For e.g. S = [2,4,5,7,9], n = 20:
![Page 36: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/36.jpg)
Fast Reconstruction for Q3
For e.g. S = [2,4,5,7,9], n = 20:
dist (P4,P
5) = 8
dist (P7,P
9) = 4
![Page 37: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/37.jpg)
Fast Reconstruction for Q3
For e.g. S = [2,4,5,7,9], n = 20:
dist (P4,P
5) = 8
dist (P7,P
9) = 4
v2
v7 v
9
v
4
T = {P2,P
4,P
5,P
7,P
9}
3 3
v4
v5
3
33
6 8
6 8
*Some edges in the above graph are ignored for visual clarity.
![Page 38: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/38.jpg)
Fast Reconstruction for Q3
Even better computation plan can be achieved by:➢ Finding a minimum-weight Steiner tree on G
➢ Introduce intermediate vertices; i.e. intermediate values.➢ Trade-off between number of aggregated keys and
reconstruction time:➢ Split S into several subqueries, issuing one key for each query.➢ The splitting is done using single-linkage clustering method. ➢ The distance betwee two “clusters” Sa and Sb are total number
of multiplications required to reconstruct samples in the union cluster Sa S∪ b.
![Page 39: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/39.jpg)
Experiments
Figure 1: Reconstruction time for Q1 & Q2.
![Page 40: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/40.jpg)
Experiments
Figure 1: Reconstruction time for Q1 & Q2.
90x speedups
![Page 41: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/41.jpg)
Experiments
Figure 2: Reconstruction time for Q3. MST(o) indicates the computation plan constructed with o intermediate values. m is the size of query result.
![Page 42: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/42.jpg)
Experiments
Figure 2: Reconstruction time for Q3. MST(o) indicates the computation plan constructed with o intermediate values. m is the size of query result.
8x speedups
![Page 43: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/43.jpg)
Experiments
Figure 3: Trade-off between number of aggregated keys and reconstruction time for Q3. k is number of sub-queries, m is the size of query result.
![Page 44: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/44.jpg)
Experiments
Figure 3: Trade-off between number of aggregated keys and reconstruction time for Q3. k is number of sub-queries, m is the size of query result.
19x speedups by splitting into 16 sub-queries.
![Page 45: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/45.jpg)
Related Works
➢ Key sharing with hierarchical structures (e.g. trees) (Tzeng '02, Benaloh '09, Atallah '09)
➢ Not applicable for multi-dimensional data not following hierarchical structure.
➢ Key Policy – Attribute based Encryption (Chase '06, Hohenberger '08, Lewko '09)
➢ Prohibitive performance overhead.
➢ Complex queries over encrypted data (Boneh '07, Shi '07)➢ Irrelevant security requirement (e.g. secrecy of all attributes).
➢ KAC follow-ups (Tong '13, Deng '14)➢ Did not address the fast reconstruction techniques.
![Page 46: NUS Computing - Privacy-Preserving Sensor Cloudchangec/publications/2016... · 2017-04-12 · Sensors are spatially arranged. Sensors continuously sense, encrypt and stream samples](https://reader035.vdocuments.net/reader035/viewer/2022070908/5f87023a785c1d49b93d57b3/html5/thumbnails/46.jpg)
Conclusions
➢ Fast reconstruction techniques for KAC enables scalable sharings of sensitive data.
➢ Our observation is also applicable to other cryptographic primitives involving group multiplications such as broadcast encryption and redactable signatures.