docs: privacy strategy document #343

EazyAl · 2023-08-17T10:01:59Z

No description provided.

docs/source/additional/privacy-strategy.rst

Signed-off-by: EazyAl <[email protected]>

SdgJlbl · 2023-08-29T06:22:00Z

I think there was a problem during the rebase, some commits are now duplicated between main and your branch.

SdgJlbl · 2023-08-29T06:26:57Z

docs/source/index.rst

@@ -128,3 +128,4 @@ Some quick links:
   additional/release.rst
   additional/faq.rst
   additional/glossary.rst
+   additional/privacy-strategy.rst


I think I would personally rather put it under the "What is Substra" section, rather than hidden at the bottom. But happy to be challenged.

I think this is quite specific and not key to what substra is - not everyone who uses substra would be keen to read this

Signed-off-by: EazyAl <[email protected]>

SdgJlbl

Thanks a lot for the PR.

The URL and some list items are not displaying correctly, I've made some suggestion.
You can check the rendering here https://owkin-substra-documentation--343.com.readthedocs.build/en/343/additional/privacy-strategy.html

Please don't forget to add a line in the changelog :)

docs/source/additional/privacy-strategy.rst

SdgJlbl · 2023-08-29T16:04:54Z

docs/source/additional/privacy-strategy.rst

+
+Here we elaborate on what PETs are and provide a quick summary of the most popular ones. We then take a look at the different types of attacks that can result in data leakage or data theft and then also propose best practices for project governance and security in order to mitigate the risks of potential attacks. We also elaborate on the collaboration required between different personas to ensure data integrity and safe handling of data. 
+
+As pointed out by Katharine Jarmul in her (primer on Privacy Enhancing Technologies)[https://martinfowler.com/articles/intro-pet.html], “Privacy is a technical, legal, political, social and individual concept.” As such, technical solutions are an important part of the answer, but they must be used in conjunction with good data governance and reliable security measures.


Suggested change

As pointed out by Katharine Jarmul in her (primer on Privacy Enhancing Technologies)[https://martinfowler.com/articles/intro-pet.html], “Privacy is a technical, legal, political, social and individual concept.” As such, technical solutions are an important part of the answer, but they must be used in conjunction with good data governance and reliable security measures.

As pointed out by Katharine Jarmul in her [primer on Privacy Enhancing Technologies](https://martinfowler.com/articles/intro-pet.html), “Privacy is a technical, legal, political, social and individual concept.” As such, technical solutions are an important part of the answer, but they must be used in conjunction with good data governance and reliable security measures.

Well, of course, it's RST 🤦
So we need the stupid syntax for external URLS:

`primer on Privacy Enhancing Technologies <https://martinfowler.com/articles/intro-pet.html>`_

(yes, backticks + a trailing underscore)

docs/source/additional/privacy-strategy.rst

Signed-off-by: EazyAl <[email protected]>

EazyAl · 2023-08-30T13:10:33Z

I have asked for review - @SdgJlbl , can we still add co-contributors?

SdgJlbl · 2023-08-31T12:30:21Z

I have asked for review - @SdgJlbl , can we still add co-contributors?

Yes, we will add the co-authors on the squash commit during the merge I think

SdgJlbl · 2023-08-31T12:41:05Z

CHANGELOG.md

+- New page added on Substra Privacy Strategy based on research by Privacy Task Force at Owkin.



Suggested change

- New page added on Substra Privacy Strategy based on research by Privacy Task Force at Owkin.

- New page added on Substra Privacy Strategy based on research by Privacy Task Force at Owkin([#343](https://github.com/Substra/substra-documentation/pull/343))

I think you're missing a space between Owkin and the opening bracket :x

SdgJlbl · 2023-08-31T12:41:30Z

docs/source/additional/privacy-strategy.rst

+We touch on a **few** of the main technologies that are making collaborative data sharing possible today in ways that can be considered more secure.
+
+**Federated Learning**:
+Federated Learning allows Machine Learning models to be sent to servers where they can train and test on data without having to ever move the data from its original location. This idea however is not restricted to machine learning and is then referred to as Federated Analytics. The Substra software enables both Federated Learning and Federated Analytics.


Suggested change

Federated Learning allows Machine Learning models to be sent to servers where they can train and test on data without having to ever move the data from its original location. This idea however is not restricted to machine learning and is then referred to as Federated Analytics. The Substra software enables both Federated Learning and Federated Analytics.

Federated Learning allows Machine Learning models to be sent to servers where they can train and test on data without having to ever move the data from its original location. This idea however is not restricted to machine learning and is then referred to as Federated Analytics. Substra enables both Federated Learning and Federated Analytics.

SdgJlbl · 2023-08-31T12:46:04Z

docs/source/additional/privacy-strategy.rst

+#. 3. The external world (outside of the network) contains malicious actors. We make no assumptions about any external communication and we aim at limiting as much as possible our exposure to the outside world.
+#. 4. Models are accessible by data scientists in the network (with the right permissions). The data scientist is responsible for making sure that the trained model exported does not contain sensitive information enabling, for example, membership attacks. (explained below)
+#. 5. Every organization in the network is a responsible actor. Every organization hosts its own node of the Substra network, and is responsible for ensuring minimal securitization of their infrastructure. Regular security audits and / or certifications are recommended.
+#. 6. In this document the focus is on protecting data rather than models — thus we do not cover Byzantine attacks [Fang, M., Cao, X., Jia, J., & Gong, N. (2020). Local model poisoning attacks to {Byzantine-Robust} federated learning]  and backdoor attacks [Bagdasaryan, E., Veit, A., Hua, Y., Estrin, D., & Shmatikov, V. (2020, June). How to backdoor federated learning.].- which are in a category of attacks that affect the quality of the generated model as opposed to compromising the data. 


Formatting of the bibliographic references

EazyAl changed the title ~~Privacy ali~~ Privacy Strategy Document Aug 17, 2023

EazyAl force-pushed the privacy-ali branch from dc50b41 to 525f5b7 Compare August 17, 2023 10:03

EazyAl changed the title ~~Privacy Strategy Document~~ doc: Privacy Strategy Document Aug 18, 2023

IC-1101asterisk reviewed Aug 21, 2023

View reviewed changes

docs/source/additional/privacy-strategy.rst Outdated Show resolved Hide resolved

EazyAl force-pushed the privacy-ali branch from e07d024 to 1aa8365 Compare August 28, 2023 15:08

EazyAl changed the title ~~doc: Privacy Strategy Document~~ doc: privacy strategy document Aug 28, 2023

EazyAl added 4 commits August 28, 2023 17:13

break test

3a76a58

Signed-off-by: EazyAl <[email protected]>

break test 2

6f0df79

Signed-off-by: EazyAl <[email protected]>

break test 2

ca53184

Signed-off-by: EazyAl <[email protected]>

Privacy Strategy Page

5ee5f1b

Signed-off-by: EazyAl <[email protected]>

EazyAl force-pushed the privacy-ali branch from a812889 to a86e77b Compare August 28, 2023 15:13

SdgJlbl reviewed Aug 29, 2023

View reviewed changes

EazyAl added 5 commits August 29, 2023 15:12

WIP

0fce335

Signed-off-by: EazyAl <[email protected]>

Fixed errors

1300247

Signed-off-by: EazyAl <[email protected]>

Fixed errors again

005a0a4

Signed-off-by: EazyAl <[email protected]>

thanks leo-paul

d4f05be

Signed-off-by: EazyAl <[email protected]>

markdown fix

9cedb7e

Signed-off-by: EazyAl <[email protected]>

EazyAl force-pushed the privacy-ali branch from 11113d3 to 9cedb7e Compare August 29, 2023 13:12

markdown fixes again

21a6622

Signed-off-by: EazyAl <[email protected]>

SdgJlbl reviewed Aug 29, 2023

View reviewed changes

EazyAl changed the title ~~doc: privacy strategy document~~ docs: privacy strategy document Aug 30, 2023

URLs fixed

8dbdb52

Signed-off-by: EazyAl <[email protected]>

EazyAl marked this pull request as ready for review August 30, 2023 13:09

SdgJlbl reviewed Aug 31, 2023

View reviewed changes

ThibaultFy closed this Sep 4, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

docs: privacy strategy document #343

docs: privacy strategy document #343

EazyAl commented Aug 17, 2023

SdgJlbl commented Aug 29, 2023

SdgJlbl Aug 29, 2023

EazyAl Aug 30, 2023

SdgJlbl left a comment

SdgJlbl Aug 29, 2023

SdgJlbl Aug 31, 2023 •

edited

Loading

SdgJlbl Aug 31, 2023

EazyAl commented Aug 30, 2023

SdgJlbl commented Aug 31, 2023

SdgJlbl Aug 31, 2023

oleobal Aug 31, 2023

SdgJlbl Aug 31, 2023

SdgJlbl Aug 31, 2023


		Here we elaborate on what PETs are and provide a quick summary of the most popular ones. We then take a look at the different types of attacks that can result in data leakage or data theft and then also propose best practices for project governance and security in order to mitigate the risks of potential attacks. We also elaborate on the collaboration required between different personas to ensure data integrity and safe handling of data.

		As pointed out by Katharine Jarmul in her (primer on Privacy Enhancing Technologies)[https://martinfowler.com/articles/intro-pet.html], “Privacy is a technical, legal, political, social and individual concept.” As such, technical solutions are an important part of the answer, but they must be used in conjunction with good data governance and reliable security measures.

		- New page added on Substra Privacy Strategy based on research by Privacy Task Force at Owkin.

	Federated Learning allows Machine Learning models to be sent to servers where they can train and test on data without having to ever move the data from its original location. This idea however is not restricted to machine learning and is then referred to as Federated Analytics. The Substra software enables both Federated Learning and Federated Analytics.
	Federated Learning allows Machine Learning models to be sent to servers where they can train and test on data without having to ever move the data from its original location. This idea however is not restricted to machine learning and is then referred to as Federated Analytics. Substra enables both Federated Learning and Federated Analytics.

docs: privacy strategy document #343

docs: privacy strategy document #343

Conversation

EazyAl commented Aug 17, 2023

SdgJlbl commented Aug 29, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SdgJlbl left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SdgJlbl Aug 31, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

EazyAl commented Aug 30, 2023

SdgJlbl commented Aug 31, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

SdgJlbl Aug 31, 2023 •

edited

Loading