8000 GitHub - wanghuazhong/ppml_IoT: Privacy preserving machine learning in the IoT domain
[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Skip to content

wanghuazhong/ppml_IoT

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

#Privacy preserving IoT: on the value of data

Concept

We do believe there is a big value in IoT data, in particular when such data are related to personal information (e.g on health or user behaviors). Such value is often "hidden" and becomes clear only after some information retrival and/or data mining algorithm and/or machine learning algorithm runs on it.

Nowadays, the extraction of useful insights from personal data is demanded to trusted third parties (see picture below). To do that, our data has to leave our private domain, and consequently we loose their control.

We hope for a future, in which the data will always be under the control of their owner, that will be able to valorize them in the most convenient way.

To support our vision, some technologies have to be developed and/or integrated and employed.

  1. A technology to track data and their usage
  2. A technology to quantify the quality/value of the data
  3. A technology to valorize data
  4. A technology to extract information from data in a privacy preserving way

Notice that, if we will be able to develop point 4 (see below for a preliminary SOTA) this will also help in point 2, at least in some contexts. As an example, let's assume to have a technology to build privacy preserving prediction models and let's assume that a user U participated to the creation of a model M with its data D. The accuracy of predictions based on M, on D and on new data, could be potentially used as a proxy to establish the value of D itself. As much is the prediction accuracy of M, as much is the quality of D and thus their potential value.

Finally observe that blockchain, seems to be an effective solution to handle at least some aspects of point 1 and 3.

Alt text

SOTA on technologies to extract information from data in a privacy preserving way

Inspiring paper by Vitaly Shmatikov on Privacy-Preserving Deep Learning

Ciao

Efficient Algorithms for Public-Private Social Networks Efficient Algorithms for Public-Private Social Networks

Privacy Preserving Clustering

Deep Learning in Python

Run Keras on Jupyter over Docker Keras on Jupyter over Docker

docker run -d -p 8888:8888 -v '/home/andrea/Documents/papers/IoTBC/Deep Learning/notebook':/notebook -e KERAS_BACKEND=tensorflow ermaker/keras-jupyter

docker run -v '/home/andrea/Documents/papers/IoTBC/Deep Learning/notebook':/notebook --rm -it ermaker/keras

Distribute Private Keys to the nodes

Inside PythonScript folder, the available script distributes a private (signing) key to a receiving node. Then, it parses each line of data tryin to verify the signature. Inside node folder, there is the project for the node. It works as follows:

1- Wait to receive a private (signing) key from the serial line

2- Increment a counter each second and send a message with the following format: (message, digest, signature)

About

Privacy preserving machine learning in the IoT domain

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C 68.0%
  • Assembly 27.7%
  • Objective-C 2.7%
  • Python 1.1%
  • Makefile 0.5%
0