[AAAI 2026] AEGIS: An Industrial Anomaly Detection System That Evolves With Field Experts - LG AI Research BLOG

Anomaly detection in manufacturing sites faces the challenge of defining a standard for 'normal' due to environmental changes and complex domain knowledge. Existing AI models are vulnerable to environmental changes, and a cognitive gap problem often arises where experts cannot understand the basis for AI's judgments.

AEGIS, unveiled by LG AI Research at AAAI 2026, is an innovative platform where AI and domain experts collaborate to manage quality. AEGIS solves these problems through three core elements.

• Explainable AI (XAI): Utilizing LLM-based agents to analyze historical records and domain documents, providing logical grounds in natural language for why a given data point was judged to be an anomaly.
• Intuitive Feedback Interface: Helps experts focus on samples with high uncertainty, and reflects labeling rules entered in natural language through a chat interface as structured data.
• Incremental Learning: Automatically monitors changes in data distribution and notifies experts, then retrains the model upon expert approval to optimize for the changing environment.

AEGIS aims to be more than a simple detection tool—it aspires to be an intelligent platform that learns from changes in the field and grows together with it.