Skip to content

Commit

Permalink
deploy: b411ae4
Browse files Browse the repository at this point in the history
  • Loading branch information
@hsiangjenli
hsiangjenli committed Nov 11, 2024
1 parent 1c3ab82 commit d3a9334
Show file tree
Hide file tree
Showing 12 changed files with 457 additions and 26 deletions.
105 changes: 102 additions & 3 deletions _sources/paper/xu2021privacy-outline.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -60,32 +60,131 @@
> - 🟥 third-party CF
#### Model Serving Phase

##### Typical types of privacy leakage
1. Disclosure of Membership
1. Class Representative
1. Property

> Guarantee on model serving phase, there are four roles:
> 1. Data Producer
> 1. Local CF
> 1. Third-party CF
> 1. Model Consumer
##### Global Model Privacy Guarantee

> - 🟩 Local CF
> - 🟩 Third-Party CF
> - 🔴 Model Consumer
#### Full Privacy Guarantee

---

## Technical {U}tility in PPML
> 1. How to release/publish data without revealing sensitive information
> 1. How the data used in the model training
> 1. The architecture of ML system prevents the disclosure of sensitive information
> The impact of using PPML can be classified into n categories:
> 1. Model Utility
> 1. Computation Utility
> 1. Communication Utility
> 1. Scalability Utility
> 1. Scenario Utility
> 1. Privacy Utility
<!-- > 1. Computation Utility
> 1. Communication Utility
> 1. Model Utility
> 1. Scalability Utility -->
### Data Publish Approach

#### Elimitation-based Approach
> - Totally eliminate the identifier from the data
> - Partially conceal quasi-identifiers
1. $k$-anonymity
1. $l$-diversity
1. $t$-closeness

#### Perturbation-based Approach
1. Differential Privacy
1. Sketching

#### Confusion-based Approach (Cryptography)
1. Symmetric encryption (e.g. AES) + Garbled circuits / Oblivious transfer
1. Homomorphic encryption
1. Functional encryption

### Data Processing Approach

#### Ordinary Computation
> If using traditional anonymization techniques or perturbation-based techniques, the identifiers are removed from the data, then can use ordinary computation
##### Elimination-based Approach
##### Perturbation-based Approach

#### Secure Computation

> Adversary setting is semi-honest or malicious
##### Additive Mask based Approach
> - Private data are masked with randomize values
> - A light-weight approach of secure computation
1. Pairwise Additive Masking Based Secure Computation Approach
- $t$-of-$n$ secret sharing
- Multi-Secret Sharing
1. Double Masking Pairwise Based Protocol
- Prevents the failure happened in Pairwise Additive Masking
1. Anonymous Communication
- DC-nets (Dining Cryptographers Networks)
- mix-nets


##### Garbled Circuit based Approach

##### Advanced Cryptography based Approach
###### Homomorphic Encryption
###### Functional Encryption

##### Mixed Protocol Approach
> Combine the above approaches
1. TASTY
1. ABY
1. ABY$^3$
1. CrypTen
1. Falcon

##### Trusted Execution Environment Approach
1. Code Authentication
1. Runtime State Integrity
1. Confidentiality

### Architectureal Appraoch

#### Delegation-based ML Architecture
> Provide computation-limited parties the capability to create and use ML models
1. CryptoML
1. SecureML

#### Distributed Selective SGD Architecture

#### Federated Learning Architecture

#### Knowledge Transfer Architecture

1. Knowledge Distillation
1. Model Compression
1. Transfer Learning

### Hybrid Approach

---
<!-- ---
## Challenge and Potential Directions
## Challenge and Potential Directions -->


<!-- References
Expand Down
7 changes: 7 additions & 0 deletions _sources/paper/xu2021privacy.rst
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,13 @@ WIP - :title-ref:`xu2021privacy`
* Authors : :cite:authors:`xu2021privacy`


An overview of mindmap
----------------------

.. raw:: html
/_static/image/image.png
<a href="_static/html/xu2021privacy.html">testurl</a>

Before starting
---------------

Expand Down
4 changes: 4 additions & 0 deletions _sources/resource/keyterm/key.rst
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,10 @@ KeyTerms
- Software As A Service
- Todo.

* - TEE
- Trusted Execution Environment
- Todo.

* - V2X
- Vehicle To Everything
- Todo.
Expand Down
4 changes: 4 additions & 0 deletions _sources/resource/ppml.rst
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,10 @@ Privacy Preserving Machine Learning framework
* https://github.com/facebookresearch/CrypTen


Others
------
* https://github.com/mpc-msri/EzPC :cite:`rathee2021sirnn`

References
----------
.. bibliography::
Expand Down
66 changes: 66 additions & 0 deletions _static/html/xu2021privacy.html
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,66 @@
<!DOCTYPE html>
<html>
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<meta http-equiv="X-UA-Compatible" content="ie=edge">
<title>Markmap</title>
<style>
* {
margin: 0;
padding: 0;
}
#mindmap {
display: block;
width: 100vw;
height: 100vh;
}
</style>
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css"><link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/style.css">
</head>
<body>
<svg id="mindmap"></svg>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/d3.min.js"></script><script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/browser/index.js"></script><script>((getMarkmap) => {
window.WebFontConfig = {
custom: {
families: [
"KaTeX_AMS",
"KaTeX_Caligraphic:n4,n7",
"KaTeX_Fraktur:n4,n7",
"KaTeX_Main:n4,n7,i4,i7",
"KaTeX_Math:i4,i7",
"KaTeX_Script",
"KaTeX_SansSerif:n4,n7,i4",
"KaTeX_Size1",
"KaTeX_Size2",
"KaTeX_Size3",
"KaTeX_Size4",
"KaTeX_Typewriter"
]
},
active: () => {
getMarkmap().refreshHook.call();
}
};
})(() => window.markmap)</script><script src="https://cdn.jsdelivr.net/npm/[email protected]/webfontloader.js" defer></script><script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/index.js"></script><script>(r => {
setTimeout(r);
})(() => {
const {
markmap,
mm
} = window;
const {
el
} = markmap.Toolbar.create(mm);
el.setAttribute('style', 'position:absolute;bottom:20px;right:20px');
document.body.append(el);
})</script><script>((getMarkmap, getOptions, root2, jsonOptions) => {
const markmap = getMarkmap();
window.mm = markmap.Markmap.create(
"svg#mindmap",
(getOptions || markmap.deriveOptions)(jsonOptions),
root2
);
})(() => window.markmap,null,{"content":"WIP - Outline of {title-ref}<code>xu2021privacy</code>","children":[{"content":"Privacy Preserving {P}hase in PPML","children":[{"content":"Model Generation","children":[],"payload":{"lines":"13,14"}},{"content":"Model Serving","children":[],"payload":{"lines":"19,20"}},{"content":"Full Pipeline","children":[],"payload":{"lines":"26,27"}}],"payload":{"lines":"5,6"}},{"content":"Privacy {G}uarantees in PPML","children":[{"content":"Object-Oriented Privacy Guarantee","children":[{"content":"Data Oriented","children":[],"payload":{"lines":"33,34"}},{"content":"Model Oriented","children":[],"payload":{"lines":"37,38"}}],"payload":{"lines":"32,33"}},{"content":"Pipeline-Orientied Privacy Guarantee","children":[{"content":"Model Generation Phase","children":[{"content":"Vanilla Local Privacy Guarantee","children":[],"payload":{"lines":"52,53"}},{"content":"Primary Local Privacy Guarantee","children":[],"payload":{"lines":"55,56"}},{"content":"Enhanced Local Privacy Guarantee","children":[],"payload":{"lines":"58,59"}}],"payload":{"lines":"46,47"}},{"content":"Model Serving Phase","children":[{"content":"Typical types of privacy leakage","children":[{"content":"Disclosure of Membership","children":[],"payload":{"lines":"64,65"}},{"content":"Class Representative","children":[],"payload":{"lines":"65,66"}},{"content":"Property","children":[],"payload":{"lines":"66,68"}}],"payload":{"lines":"63,64"}},{"content":"Global Model Privacy Guarantee","children":[],"payload":{"lines":"74,75"}}],"payload":{"lines":"61,62"}},{"content":"Full Privacy Guarantee","children":[],"payload":{"lines":"79,80"}}],"payload":{"lines":"40,41"}}],"payload":{"lines":"30,31"}},{"content":"Technical {U}tility in PPML","children":[{"content":"Data Publish Approach","children":[{"content":"Elimitation-based Approach","children":[{"content":"<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><mi>k</mi></mrow><annotation encoding=\"application/x-tex\">k</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.69444em;vertical-align:0em;\"></span><span class=\"mord mathnormal\" style=\"margin-right:0.03148em;\">k</span></span></span></span>-anonymity","children":[],"payload":{"lines":"106,107"}},{"content":"<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><mi>l</mi></mrow><annotation encoding=\"application/x-tex\">l</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.69444em;vertical-align:0em;\"></span><span class=\"mord mathnormal\" style=\"margin-right:0.01968em;\">l</span></span></span></span>-diversity","children":[],"payload":{"lines":"107,108"}},{"content":"<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><mi>t</mi></mrow><annotation encoding=\"application/x-tex\">t</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.61508em;vertical-align:0em;\"></span><span class=\"mord mathnormal\">t</span></span></span></span>-closeness","children":[],"payload":{"lines":"108,110"}}],"payload":{"lines":"102,103"}},{"content":"Perturbation-based Approach","children":[{"content":"Differential Privacy","children":[],"payload":{"lines":"111,112"}},{"content":"Sketching","children":[],"payload":{"lines":"112,114"}}],"payload":{"lines":"110,111"}},{"content":"Confusion-based Approach (Cryptography)","children":[{"content":"Symmetric encryption (e.g. AES) + Garbled circuits / Oblivious transfer","children":[],"payload":{"lines":"115,116"}},{"content":"Homomorphic encryption","children":[],"payload":{"lines":"116,117"}},{"content":"Functional encryption","children":[],"payload":{"lines":"117,119"}}],"payload":{"lines":"114,115"}}],"payload":{"lines":"100,101"}},{"content":"Data Processing Approach","children":[{"content":"Ordinary Computation","children":[{"content":"Elimination-based Approach","children":[],"payload":{"lines":"124,125"}},{"content":"Perturbation-based Approach","children":[],"payload":{"lines":"125,126"}}],"payload":{"lines":"121,122"}},{"content":"Secure Computation","children":[{"content":"Additive Mask based Approach","children":[{"content":"Pairwise Additive Masking Based Secure Computation Approach","children":[{"content":"<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><mi>t</mi></mrow><annotation encoding=\"application/x-tex\">t</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.61508em;vertical-align:0em;\"></span><span class=\"mord mathnormal\">t</span></span></span></span>-of-<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><mi>n</mi></mrow><annotation encoding=\"application/x-tex\">n</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.43056em;vertical-align:0em;\"></span><span class=\"mord mathnormal\">n</span></span></span></span> secret sharing","children":[],"payload":{"lines":"136,137"}},{"content":"Multi-Secret Sharing","children":[],"payload":{"lines":"137,138"}}],"payload":{"lines":"135,138"}},{"content":"Double Masking Pairwise Based Protocol","children":[{"content":"Prevents the failure happened in Pairwise Additive Masking","children":[],"payload":{"lines":"139,140"}}],"payload":{"lines":"138,140"}},{"content":"Anonymous Communication","children":[{"content":"DC-nets (Dining Cryptographers Networks)","children":[],"payload":{"lines":"141,142"}},{"content":"mix-nets","children":[],"payload":{"lines":"142,145"}}],"payload":{"lines":"140,145"}}],"payload":{"lines":"131,132"}},{"content":"Garbled Circuit based Approach","children":[],"payload":{"lines":"145,146"}},{"content":"Advanced Cryptography based Approach","children":[{"content":"Homomorphic Encryption","children":[],"payload":{"lines":"148,149"}},{"content":"Functional Encryption","children":[],"payload":{"lines":"149,150"}}],"payload":{"lines":"147,148"}},{"content":"Mixed Protocol Approach","children":[{"content":"TASTY","children":[],"payload":{"lines":"154,155"}},{"content":"ABY","children":[],"payload":{"lines":"155,156"}},{"content":"ABY<span class=\"katex\"><span class=\"katex-mathml\"><math xmlns=\"http://www.w3.org/1998/Math/MathML\"><semantics><mrow><msup><mrow></mrow><mn>3</mn></msup></mrow><annotation encoding=\"application/x-tex\">^3</annotation></semantics></math></span><span class=\"katex-html\" aria-hidden=\"true\"><span class=\"base\"><span class=\"strut\" style=\"height:0.8141079999999999em;vertical-align:0em;\"></span><span class=\"mord\"><span></span><span class=\"msupsub\"><span class=\"vlist-t\"><span class=\"vlist-r\"><span class=\"vlist\" style=\"height:0.8141079999999999em;\"><span style=\"top:-3.063em;margin-right:0.05em;\"><span class=\"pstrut\" style=\"height:2.7em;\"></span><span class=\"sizing reset-size6 size3 mtight\"><span class=\"mord mtight\">3</span></span></span></span></span></span></span></span></span></span></span>","children":[],"payload":{"lines":"156,157"}},{"content":"CrypTen","children":[],"payload":{"lines":"157,158"}},{"content":"Falcon","children":[],"payload":{"lines":"158,160"}}],"payload":{"lines":"151,152"}},{"content":"Trusted Execution Environment Approach","children":[{"content":"Code Authentication","children":[],"payload":{"lines":"161,162"}},{"content":"Runtime State Integrity","children":[],"payload":{"lines":"162,163"}},{"content":"Confidentiality","children":[],"payload":{"lines":"163,165"}}],"payload":{"lines":"160,161"}}],"payload":{"lines":"127,128"}}],"payload":{"lines":"119,120"}},{"content":"Architectureal Appraoch","children":[{"content":"Delegation-based ML Architecture","children":[{"content":"CryptoML","children":[],"payload":{"lines":"169,170"}},{"content":"SecureML","children":[],"payload":{"lines":"170,172"}}],"payload":{"lines":"167,168"}},{"content":"Distributed Selective SGD Architecture","children":[],"payload":{"lines":"172,173"}},{"content":"Federated Learning Architecture","children":[],"payload":{"lines":"174,175"}},{"content":"Knowledge Transfer Architecture","children":[{"content":"Knowledge Distillation","children":[],"payload":{"lines":"178,179"}},{"content":"Model Compression","children":[],"payload":{"lines":"179,180"}},{"content":"Transfer Learning","children":[],"payload":{"lines":"180,182"}}],"payload":{"lines":"176,177"}}],"payload":{"lines":"165,166"}},{"content":"Hybrid Approach","children":[],"payload":{"lines":"182,183"}}],"payload":{"lines":"83,84"}}],"payload":{"lines":"0,1"}},{})</script>
</body>
</html>
Binary file modified objects.inv
View file
Binary file not shown.
4 changes: 2 additions & 2 deletions paper/gill2022ai.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -369,10 +369,10 @@ <h2>Key Concepts and Terminologies<a class="headerlink" href="#key-concepts-and-
</section>
<section id="proposed-methodologies">
<h2>Proposed Methodologies<a class="headerlink" href="#proposed-methodologies" title="Link to this heading">#</a></h2>
<figure class="align-default" id="id42">
<figure class="align-default" id="id43">
<img alt="../_images/image-1.png" src="../_images/image-1.png" />
<figcaption>
<p><span class="caption-text">This figure is taken from the paper <span id="id4">[<a class="reference internal" href="#id40" title="Sukhpal Singh Gill, Minxian Xu, Carlo Ottaviani, Panos Patros, Rami Bahsoon, Arash Shaghaghi, Muhammed Golec, Vlado Stankovski, Huaming Wu, Ajith Abraham, and others. Ai for next generation computing: emerging trends and future directions. Internet of Things, 19:100514, 2022.">1</a>]</span></span><a class="headerlink" href="#id42" title="Link to this image">#</a></p>
<p><span class="caption-text">This figure is taken from the paper <span id="id4">[<a class="reference internal" href="#id40" title="Sukhpal Singh Gill, Minxian Xu, Carlo Ottaviani, Panos Patros, Rami Bahsoon, Arash Shaghaghi, Muhammed Golec, Vlado Stankovski, Huaming Wu, Ajith Abraham, and others. Ai for next generation computing: emerging trends and future directions. Internet of Things, 19:100514, 2022.">1</a>]</span></span><a class="headerlink" href="#id43" title="Link to this image">#</a></p>
</figcaption>
</figure>
<ul class="simple">
Expand Down
Loading

0 comments on commit d3a9334

Please sign in to comment.