Learning System Reliability under Uncertainty is a research topic that investigates when a learning system can be considered reliable and under what conditions it structurally collapses in the presence of uncertainty across data, model architecture, and the training process. This study moves beyond conventional average performance–centric evaluation, focusing on reliability as a core objective that encompasses stability, reproducibility, and robustness. In particular, it aims to systematically define uncertainty conditions and to establish an integrated framework that explains the mechanisms of reliability collapse and enables verifiable evaluation.

• Problem Reliability degradation and structural collapse of learning systems under uncertainty.
• Perspective Reliability evaluation beyond average performance (stability, reproducibility, robustness).
• Framework Conditions–mechanisms–validation-based analytical structure
• Method Systematic uncertainty modeling across data, architecture, and training with boundary analysis.
• Goal Establishing generalizable analysis and evaluation criteria for high-reliability learning systems.