Table 3 Ablation study evaluating the impact of encoder type, contrastive strategy, and training constraints on ACCEPT performance. \(^*\)Compare queue sampling using random negatives and hard negatives (lowest cosine similarity) to evaluate how sensitive the contrastive alignment is to negative sample selection.
From: Diagnostic forecasting of battery degradation through contrastive learning
Ablation | MSE | MAE | MAPE |
|---|---|---|---|
Full model (baseline) | 0.0005 | 0.0082 | 0.0088 |
Conv \(\rightarrow\) Transformer (physical encoder) | 0.0012 | 0.0372 | 0.0388 |
No queue mechanism | 0.0009 | 0.0104 | 0.0110 |
Random vs. hard negative sampling\(^*\) | 0.0016 | 0.0189 | 0.0362 |
