2020
DOI: 10.1016/j.pmcj.2020.101132
| View full text |Cite
|
Sign up to set email alerts
|

Abstract: Sensitive inferences and user re-identification are major threats to privacy when raw sensor data from wearable or portable devices are shared with cloud-assisted applications. To mitigate these threats, we propose mechanisms to transform sensor data before sharing them with applications running on users' devices. These transformations aim at eliminating patterns that can be used for user reidentification or for inferring potentially sensitive activities, while introducing a minor utility loss for the target a… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
14
0

Year Published

2020
2020
2024
2024

Publication Types

Select...
4
3
2

Relationship

3
6

Authors

Journals

citations
Cited by 27 publications
(14 citation statements)
references
References 38 publications
0
14
0
Order By: Relevance
“…It aims at the trained systems that are deployed to offer Inference-as-a-Service [145]. Most methods in private inference are similar to those in private training, except for the Information-Theoretic privacy [138,137,139]. It is typically used to offer information-theoretic mathematical or empirical evidence of how these methods operate to improve privacy.…”
Section: Privacy Preservingmentioning
confidence: 99%
“…It aims at the trained systems that are deployed to offer Inference-as-a-Service [145]. Most methods in private inference are similar to those in private training, except for the Information-Theoretic privacy [138,137,139]. It is typically used to offer information-theoretic mathematical or empirical evidence of how these methods operate to improve privacy.…”
Section: Privacy Preservingmentioning
confidence: 99%
“…In particular, we show how to generate natural-looking adversarial images either by selectively modifying colors within chosen ranges that we perceive as natural or by enhancing details in the image. The references that will be covered in the tutorial are [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18].…”
Section: Context Motivation and Descriptionmentioning
confidence: 99%
“…The architecture of a DNN is mainly defined by the type and number of the layers, and the way these layers are connected to each other [41]. A commonly used DNN architecture to classify sensor data is composed of a convolutional neural network (CNN) followed by a feedforward neural network (FNN) or a recurrent neural network (RNN) [10], [11], [12], [13], [14], [15], [16], [17], [18]. A CNN is inherently adaptive to input data of variable dimensions and learns features relevant for the downstream task.…”
Section: Related Workmentioning
confidence: 99%
“…For example, static acceleration, that shows the magnitude and direction of to the earth's gravitational force, helps to recognize the wearer's posture; whereas dynamic acceleration, that shows changes in the motion velocity of the wearer, can be mapped to the wearer's different activities [8]. The time series generated by these motion sensors are processed over temporal windows and classified by deep neural networks (DNNs) [9], [10], [11], which process sensor data with pre-defined, fixed dimensions [11], [12], [13], [14], [15], [16], [17], [18], and cannot reliably handle, at inference time, dynamic situations (e.g. when the sampling rate changes or some sensors are dropped), which are important for energy preservation [19], [20], privacy protection [21], [22] and fault tolerance [23], [24].…”
Section: Introductionmentioning
confidence: 99%