표 2. | Table 2. 각 모델별 연산량(FLOPs), 추론/학습 시간 및 파라미터 수 복잡도 비교 | Comparison of computational load (FLOPs), inference/training time, and parameter count complexity for each model.

Generating model	Computational load (MFLOPs)	Inference time (ms)	Training time (hours)
Supervised learning	1.2	≤0.05	3.5
Unsupervised learning	0.07	≤0.01	1.5
Reinforcement learning	0.9～1.1	≤0.05	≥24 (including NAS search)