[Paper Review] Towards Interpretable Federated Learning

Anran Li, Rui Liu+3 more|arXiv (Cornell University)|Feb 27, 2023

Privacy-Preserving Technologies in Data9 citations

TL;DR

This paper is the first survey of interpretable federated learning (IFL), proposing a taxonomy, analyzing representative IFL approaches, evaluation metrics, and future directions.

ABSTRACT

Federated learning (FL) enables multiple data owners to build machine learning models collaboratively without exposing their private local data. In order for FL to achieve widespread adoption, it is important to balance the need for performance, privacy-preservation and interpretability, especially in mission critical applications such as finance and healthcare. Thus, interpretable federated learning (IFL) has become an emerging topic of research attracting significant interest from the academia and the industry alike. Its interdisciplinary nature can be challenging for new researchers to pick up. In this paper, we bridge this gap by providing (to the best of our knowledge) the first survey on IFL. We propose a unique IFL taxonomy which covers relevant works enabling FL models to explain the prediction results, support model debugging, and provide insights into the contributions made by individual data owners or data samples, which in turn, is crucial for allocating rewards fairly to motivate active and reliable participation in FL. We conduct comprehensive analysis of the representative IFL approaches, the commonly adopted performance evaluation metrics, and promising directions towards building versatile IFL techniques.

Motivation & Objective

• Motivate the need for interpretable federated learning in privacy-sensitive, mission-critical applications.
• Provide a comprehensive taxonomy of IFL covering client/sample/feature selection, model optimization, and contribution evaluation.
• Analyze representative IFL approaches, evaluation metrics, and privacy considerations across FL settings.
• Identify challenges and propose directions to balance interpretability, performance, and privacy in FL.

Proposed method

• Proposes a unique IFL taxonomy grounded in stakeholder roles and privacy protections.
• Classifies IFL approaches by stages: client selection, sample selection, feature selection, model optimization, and contribution evaluation.
• Reviews interpretable techniques across model-agnostic and model-specific paradigms, including incentive-based and robust aggregation methods.
• Synthesizes evaluation metrics for effectiveness (faithfulness, post-interpretation performance) and efficiency (computation/communication costs).
• Outlines future directions like interpretable model extraction, hard sample-aware IFL, and robustness to complex threat models.

Figure 1: The proposed taxonomy for the IFL literature.

Experimental results

Research questions

• RQ1 What are the key components and stakeholders involved in interpretable federated learning?
• RQ2 What taxonomy best captures the existing IFL approaches and privacy considerations across FL stages?
• RQ3 What are the main methods for achieving interpretability in client/sample/feature selection and model optimization within FL?
• RQ4 How should IFL approaches be evaluated in terms of effectiveness and efficiency?
• RQ5 What future directions can advance interpretable, private, and robust FL systems?

Key findings

• IFL is interdisciplinary and challenging due to data invisibility and resource constraints in FL.
• A taxonomy and structured survey of IFL approaches across five stages provides a cohesive view of the field.
• Interpretable techniques span client/sample/feature selection, model construction, robust aggregation, and contribution evaluation, with both model-agnostic and model-specific methods.
• Evaluation metrics distinguish faithfulness and post-interpretation performance (effectiveness) as well as computation and communication costs (efficiency).
• Identified future directions include interpretable model extraction, hard sample-aware, privacy-preserving IFL, and addressing complex threat models.