PET privacy enhancing techniques

Referecne:  https://www.drivendata.org/competitions/98/nist-federated-learning-1/rules/

privacy-preserving federated learning (PPFL) solutions

 democracy-affirming technologies.

 the global federated model is trained, the parameters related to the local models could be used to learn about the sensitive information contained in the training data of each client. Similarly, the released global model could also be used to infer sensitive information about the training datasets used.

1.4 GOALS AND OBJECTIVES:

• Drive innovation in the technological development and application of novel privacy enhancing technologies;

• Deliver strong privacy guarantees against a set of common threats and privacy attacks; and

• Generate effective models to accomplish a set of predictive or analytical tasks that support the use cases.

Organizers seek to mature federated learning approaches and build trust in adoption by accelerating the development of efficient PPFL solutions that leverage a combination of input and output privacy techniques to:

Phase 1: Concept Paper. Blue Team Participants will produce a technical white paper (“Concept Paper” or “White Paper”) setting out their proposed solution approach. Technical papers will be evaluated by a panel of judges across a set of weighted criteria. Participants will be eligible to win prizes awarded to the top technical papers, ranked by points awarded.

As you propose your technical solutions, be prepared to clearly describe the technical approaches and sketch out proof of or justification for privacy guarantees. Participants should consider a broad range of privacy threats during the model training and model use phases and consider technical and process aspects including but not limited to cryptographic and non-cryptographic methods, and protection needed within the deployment environment.

Successful technical approaches and proofs of privacy guarantees will include the design of any algorithms, protocols, etc. utilized, as well as formal or informal arguments of how the solution will provide privacy guarantees. Participants will clearly list any additional privacy issues specific to the technological approaches used and justify initial enhancements or novelties compared to the current state-of-the-art. Participant submissions must describe how the solution will cater to the types of data provided to participants and how generalizable the solution is to multiple domains. Expected efficiency/scalability of improvements, privacy vs. utility trade off should be articulated, if possible, at this conceptual stage.

Q: what is the definition of privacy guarantee? 

a one-page abstract and a Concept Paper.

Abstract: The one-page abstract must include a title and a brief description of the proposed solution, including the proposed privacy mechanisms and architecture of the federated model. The description should also describe the proposed machine learning model and expected results with regard to accuracy. Successful abstracts will outline how solutions will achieve privacy while minimizing loss to accuracy, a proposed solution, and the anticipated results, as more fully described on the Challenge Website. Abstracts must be submitted by following the instructions on the Challenge Website. Abstracts will be screened by the DrivenData and Organizers’ staff for contest eligibility and used to ensure the composition of the judging panel’s expertise aligns to proposed solutions that will be evaluated throughout the course of the Challenge. Feedback will not be provided.
Concept Paper: The Concept Paper should conceptualize solutions that describe the technical approaches and lay out the proof of privacy guarantees that solve a set of predictive or analytic tasks that support the use cases. Successful Concept Papers will incorporate the originally submitted abstract and be no more than ten pages in length. References will not count towards page length. Participant submissions shall:

• Include a title and abstract for the solution

• Clearly articulate the selected track(s) the solution addresses, understanding of the problem, and opportunities for privacy technology within the current state-of-the-art.

• Clearly describe the technical approaches and proof of privacy guarantees based on their described threat model, including:

• The design of any algorithms, protocols, etc. utilized,

• The formal or informal arguments of how the solution will provide privacy guarantees.

• Clearly list any additional privacy issues specific to the technological approaches used.

• Justify initial enhancement or novelty compared to the state-of-the-art.

• Articulate:

• The expected efficiency and scalability of the privacy solution,

• The expected accuracy and performance of the model,

• The expected tradeoffs between privacy and accuracy/utility,

• How the explainability of model outputs may be impacted by your privacy solution,

• The feasibility of implementing the solution within the competition timeframe.

• Describe how the solution will cater to the types of data provided to participants and articulate what additional work may be needed to generalize the solution to other types of data.

• Articulate the anticipated use and purpose of licensed software.

• Be free from typographical and grammatical errors.

Participants should refer to Section 7 on general submission requirements for additional guidance and style guidelines.
Judges will score the Concept Papers against the weighted criteria outlined in the table below. Solutions will need to carefully consider trade-offs between criteria such as privacy, accuracy, and efficiency, and should take the weightings of the criteria into account when considering these trade-offs. Concept Papers must also demonstrate how acceptable levels of both privacy and accuracy will be achieved – one must not be completely traded off for the other (a fully privacy-preserving but totally inaccurate model is not of use to anyone). Proposals that do not sufficiently demonstrate how both privacy and accuracy will be achieved will not be eligible to score points in the remaining criteria.