[논문 리뷰] Transferable XAI: Relating Understanding Across Domains with Explanation Transfer

Fei Wang, Yifan Zhang외 1명|arXiv (Cornell University)|2026. 02. 14.

Explainable Artificial Intelligence (XAI)인용 수 0

한 줄 요약

본 논문은 Transferable XAI를 제안한다. 이는 관련된 AI 도메인 간 설명을 전이하기 위한 선형 변환 프레임워크이며(서브스페이스, 태스크, 속성), 건강 위험 및 대기 오염 설정에서 태스크 전이와 속성 전이로 평가된다.

ABSTRACT

Current Explainable AI (XAI) focuses on explaining a single application, but when encountering related applications, users may rely on their prior understanding from previous explanations. This leads to either overgeneralization and AI overreliance, or burdensome independent memorization. Indeed, related decision tasks can share explanatory factors, but with some notable differences; e.g., body mass index (BMI) affects the risks for heart disease and diabetes at the same rate, but chest pain is more indicative of heart disease. Similarly, models using different attributes for the same task still share signals; e.g., temperature and pressure affect air pollution but in opposite directions due to the ideal gas law. Leveraging transfer of learning, we propose Transferable XAI to enable users to transfer understanding across related domains by explaining the relationship between domain explanations using a general affine transformation framework applied to linear factor explanations. The framework supports explanation transfer across various domain types: translation for data subspace (subsuming prior work on Incremental XAI), scaling for decision task, and mapping for attributes. Focusing on task and attributes domain types, in formative and summative user studies, we investigated how well participants could understand AI decisions from one domain to another. Compared to single-domain and domain-independent explanations, Transferable XAI was the most helpful for understanding the second domain, leading to the best decision faithfulness, factor recall, and ability to relate explanations between domains. This framework contributes to improving the reusability of explanations across related AI applications by explaining factor relationships between subspaces, tasks, and attributes.

연구 동기 및 목표

• 사용자가 한 AI 도메인의 이해를 관련 도메인에 과잉 일반화나 암기 부담 없이 적용하도록 돕는다.
• 서브스페이스, 태스크, 속성 간 설명 전이를 위한 해석 가능한 affine transformation 프레임워크를 도입한다.
• 설명 신뢰성을 유지하면서 전이 매핑의 희소성을 통해 인지 부담을 줄인다.
• 실세계에 유사한 건강 위험 및 대기 오염 시나리오에서 Task Transfer와 Attributes Transfer를 형성적으로 및 총괄적으로 평가한다.

제안 방법

• 가중치 w와 상대 속성 값 χ를 사용한 선형 요인 설명으로 예측을 설명한다.
• 원 도메인 설명을 대상 도메인 설명으로 매핑하기 위해 w_T = A w_O + b의 affine transformation을 정의한다.
• 도메인 유형별 전이 구현: Subspace는 평행이동(A = I, b ≠ 0); Task는 스케일링(A = diag(κ); b = 0); Attributes는 매핑(w_T = Mχ^T w_O, 희소한 Mχ로).
• 인지 부하를 최소화하면서 충실도를 유지하기 위해 희소성 정규화를 사용해 Original 및 Target 설명기를 공동으로 학습한다(L = L_O + L_T + λ L_s).
• 건강 위험 및 대기 오염에 걸쳐 두 가지 실험(Task Transfer 및 Attributes Transfer)을 평가하여 충실도, 요인 기억, 도메인 간 관계 이해를 평가한다.

Figure 1. Concept of Transferable XAI for cross-domain understanding. a) Original Domain: Initial understanding of AI decisions. b) Target Domain: Understanding goal for the second domain. c) Mismatched Domains: Applying original understanding to the target domain results in knowledge gaps (gray) an

실험 결과

연구 질문

• RQ1 How effectively can users transfer understanding of AI decisions from an Original domain to a Target domain using Transferable XAI?
• RQ2 Does Transferable XAI improve decision faithfulness, factor recall, and the ability to relate explanations across domains compared to single-domain explanations?
• RQ3 What are the differences in cognitive load and effectiveness among Subspace, Task, and Attributes transfers?
• RQ4 Do translation, scaling, and mapping transfers support intuitive, sparse explanations for lay users across domains?

주요 결과

• Transferable XAI best helped users understand AI decisions in the Target domain compared to other explanations.
• It achieved the best decision faithfulness and factor recall in the Target domain.
• It improved users’ ability to relate explanations across domains.
• Sparsity regularization reduces cognitive load by limiting the number of transferred factors.
• The framework subsumes Incremental XAI by unifying subspace transfer with additional domain transfers.

Figure 2. Conceptual example of affine transformations across different domain types, shown with 2-dimensional data (two attributes) for simplicity. See technical details in Section 3.3 .