---

Privacy-aware Data Science

• Also known as Differential Privacy or Data Privacy, this course (unit) was originally designed for various groups of research students in some top Asia Pacific universities, including SRM Institute of Science and Technology, Hunan University, University of Chinese Academy of Sciences etc. (since 2018).

• Materials in this course include resources collected from various open-source online repositories. You are free to use, change and distribute this package.

• If you found any issue/bug for this site, please submit an issue at tulip-lab/privacy-aware-data-science:

• Pull requests are welcome:

• Preliminary unit 👉 :

• Point of Contact 👉 : Prof. Gang Li

Prepared by 🌷 TULIP Lab

---

💡 Content

This course (aka unit) offers a focused study of Differential Privacy, tailored for data science and computer science professionals. It starts with an overview of data privacy concerns, leading into the core concepts of differential privacy, including ε-differential privacy, δ-approximations, and noise addition mechanisms like the Laplace and Exponential methods.

Key components include applying differential privacy to statistical analysis and machine learning, adapting conventional techniques to uphold privacy standards. Advanced topics cover federated learning and decentralized systems, emphasizing the field's evolving nature. Discussions on ethical and legal aspects of data privacy are included, preparing students to implement privacy-preserving solutions in various professional settings. The course aims to equip participants with essential skills in designing and managing privacy-conscious data science projects.

📒 Sessions

Students will have access to a comprehensive range of subject materials, comprising slides handouts, and relevant readings. It is recommended that students commence their engagement with each session by thoroughly reviewing the pertinent slides handouts and readings to obtain a comprehensive understanding of the content.

Additionally, students are encouraged to supplement their knowledge by conducting independent research, utilizing online resources or referring to textbooks that cover relevant information related to the topics under study.

🗓️ Session Plan

The proposed unit is structured to encompass a total of 100 class hours. This allocation includes 72 hours dedicated to instruction and teaching, complemented by 28 hours set aside for student presentations and discussions.

For optimal integration into university curricula, it is suggested that this unit be divided into two distinct segments (or two consecutive units). This approach is more aligned with typical academic scheduling and facilitates a more manageable and effective learning experience.

Privacy-aware Data Science (I)

The unit plan is as below:

🔬 Session	🏷️ Category	📒 Topic	🎯 ULOs	👨‍🏫 Activity
0️⃣	Preliminary	📖 Induction	ULO1
1️⃣	Preliminary	📖 Theoretical Foundations	ULO1
2️⃣	Core	📖 Data Privacy	ULO1
3️⃣	Core	📖 Privacy Attacks	ULO1 ULO2
4️⃣	Core	📖 Differential Privacy	ULO1 ULO2
5️⃣	Core	📖 Composition of Differential Privacy	ULO1 ULO2
6️⃣	Core	📖 Sparse Vector Technique	ULO1 ULO2
7️⃣	Core	📖 Query Release and The Net Mechanism	ULO1 ULO2
🅰️	Student Work	📖 Selected Topics in DP	ULO3
8️⃣	Core	📖 DUA: Database Update Algorithm	ULO1 ULO2
9️⃣	Core	📖 PTR Mechanism and S&A Mechanism	ULO1 ULO2 ULO3
🔟	Core	📖 Fundamental Law of Information Reconstruction	ULO1 ULO2 ULO3
🅱️	Student Work	📖 Selected Topics in DP	ULO3

Privacy-aware Data Science (II)

The unit plan is as below:

🔬 Session	🏷️ Category	📒 Topic	🎯 ULOs	👨‍🏫 Activity
1️⃣	Advanced	📖 PATE	ULO1
2️⃣	Advanced	📖 M-DP and Local-DP	ULO1
3️⃣	Advanced	📖 DP Learning	ULO1 ULO2
4️⃣	Advanced	📖 DP SGD	ULO1 ULO2
5️⃣	Advanced	📖 DP Clustering	ULO1 ULO2
6️⃣	Advanced	📖 Renyi-DP and zCDP	ULO1 ULO2
7️⃣	Advanced	📖 Privacy Amplification	ULO1 ULO2
🅰️	Student Work	📖 Selected Topics in Advanced DP	ULO3
8️⃣	Advanced	📖 TBA	ULO1 ULO2
9️⃣	Advanced	📖 TBA	ULO1 ULO2 ULO3
🔟	Advanced	📖 TBA	ULO1 ULO2 ULO3
🅱️	Student Work	📖 Selected Topics in Advanced DP	ULO3
🏆	Advanced	📖 [Invited Talk and Discussions]	ULO1 ULO2

🈵 Assessment

Every cohort might be assessed differently, depending on the specific requirements of your universities.

The assessment of the unit is mainly aimed at assessing the students' achievement of the unit learning outcomes (ULOs, a.k.a. objectives), and checking the students' mastery of those theory and methods covered in the unit.

🗓️ Submission Due Dates

• HNU 2025 - The assessment can be found here, and submissions due date is 🗓️ Wednesday, 08/10/2025,
• SRM 2024 - The final assessment files submissions due date is 🗓️ Saturday, 18/05/2024 (tentative), group of one member only (individual work) for all tasks.

It is expected that you will submit each assessment component on time. You will not be allowed to start everything at the last moment, because we will provide you with feedback that you will be expected to use in future assessments.

㊙️

If you find that you are having trouble meeting your deadlines, contact the Unit Chair.

📚 References

This course uses several key references or textbooks, together with relevant publications from TULIP Lab:

• Digital Privacy Under Attack: Challenges and Enablers, by Baobao Song, Shiva Raj Pokhrel, Mengyue Deng, Qiujun Lan, Tianqing Zhu, Robin Doss, Gang Li
• The Algorithmic Foundations of Differential Privacy, by Cynthia Dwork and Aaron Roth
• Differentially Private Data Publishing and Analysis: a Survey, Tianqing Zhu, Gang Li, Wanlei Zhou, and Philip S Yu. IEEE Transactions on Knowledge and Data Engineering, 2017, 29(8): 1619-1638.
• Differential Privacy and Applications, Tianqing Zhu, Gang Li, Wanlei Zhou, Philip S. Yu. 2017. AIS 69, Springer.
• Research Publications, various resources and readings

👉 Contributors

Thanks goes to these wonderful people 🌷

Made with contributors-img.