Detect by Yourself: Self-Designing Agentic Workflows for Few-Shot Graph Anomaly Detection

Graph anomaly detection aims to identify anomaly nodes in attributed graphs and plays an important role in real-world applications. However, existing graph anomaly detection methods still face two key challenges: 1) fixed pipelines, which restrict their adaptability across different graph tasks under limited supervision; 2) weak evidence, which prevents them from explicitly incorporating contextual and structural anomaly signals into the detection process. In this paper, we propose a novel framework, self-designing agentic workflows for few-shot graph anomaly detection (SignGAD). Specifically, we propose a novel paradigm that reformulates graph anomaly detection task from training a fixed anomaly detector to designing task-conditioned detection workflows. By constructing detection workflows, SignGAD selects suitable graph encodings and detector designs to exploit task-specific anomaly evidence. Meanwhile, we introduce a guarded final refit strategy to refine the selected workflow by calibrating refit acceptance, enhancing reliability under limited supervision. Extensive experiments conducted on several real-world datasets demonstrate that SignGAD achieves strong performance against state-of-the-art methods, highlighting its effectiveness on graph anomaly detection tasks.

📋 한 줄 요약

**[Graph Anomaly Detection / Few-shot]** SignGAD가 fixed pipeline 대신 task-conditioned detection workflow를 self-design — graph encoding·detector 선택을 task별 anomaly evidence에 맞추고 guarded final refit로 reliability 보강, 여러 real-world 데이터셋 SOTA.

🎯 핵심 기여도

Graph anomaly detection이 attributed graph의 anomaly node 식별을 목표로 하며 실세계 응용에 중요하지만 두 key challenge 직면 — (1) fixed pipeline으로 limited supervision 하 다양 graph task adaptability 제한, (2) weak evidence로 contextual·structural anomaly signal을 detection process에 explicitly 통합 못함.
Self-designing agentic workflow for few-shot graph anomaly detection(SignGAD) 프레임워크 제안.
새 패러다임 — graph anomaly detection task를 fixed anomaly detector 학습이 아닌 task-conditioned detection workflow 설계로 reformulate.
Detection workflow 구성으로 task-specific anomaly evidence를 활용할 suitable graph encoding·detector design 선택; guarded final refit strategy 도입해 refit acceptance 보정·limited supervision 하 reliability 강화.

💡 핵심 아이디어

Graph anomaly detection의 fixed pipeline·weak evidence 한계는 task-conditioned workflow를 self-design하는 agentic 패러다임으로 해소 가능하며, guarded refit acceptance가 limited supervision 하 reliability를 유지하는 핵심 메커니즘이다.

🔬 기술적 접근법

**방법론**: SignGAD — self-designing agentic workflow.
**핵심 기법**: (1) Anomaly detection을 fixed detector 학습이 아닌 task-conditioned workflow 설계 문제로 reformulate, (2) Suitable graph encoding 자동 선택, (3) Task-specific anomaly evidence에 맞는 detector design 선택, (4) Guarded final refit으로 refit acceptance calibration, (5) Few-shot supervision 하 reliability 보장.

📊 주요 결과

여러 real-world 데이터셋에서 SOTA 방법 대비 strong 성능 달성.
Limited supervision 하 anomaly detection task에 효과적임을 광범위 실험 입증.
Self-designing workflow가 fixed pipeline 대비 task adaptability 우수.

💭 의의 및 한계

**의의**: Graph anomaly detection의 패러다임 — fixed detector → workflow self-design — 전환, agentic 접근의 graph task 적용 가능성 입증, few-shot 환경의 reliability 강화. **한계**: Self-design 과정의 계산 비용, agentic workflow 탐색 공간의 도메인 의존성, abstract에서 구체적 수치·데이터셋 미명시.

🚀 실용적 활용

Fraud detection·network intrusion의 few-shot anomaly detection.
다양 graph task의 자동 detector 설계.
Agentic ML 파이프라인의 graph domain 적용.