## Line Chart: NDCG@10/% vs Dimensions
### Overview
The chart compares the performance of four different methods (Original(MRL), search-adaptor, MRL-Adaptor, SMEC) across varying dimensions (128 to 3072) using the metric NDCG@10/%. All methods show upward trends, with SMEC consistently outperforming others.
### Components/Axes
- **X-axis (Dimensions)**: Logarithmic scale with values 128, 256, 512, 768, 1536, 3072.
- **Y-axis (NDCG@10/%)**: Linear scale from 50% to 62%.
- **Legend**: Located in the bottom-right corner, mapping:
- Blue circles: Original(MRL)
- Orange crosses: search-adaptor
- Green triangles: MRL-Adaptor
- Yellow squares: SMEC
### Detailed Analysis
1. **Original(MRL)** (Blue circles):
- Starts at ~49% at 128 dimensions.
- Gradually increases to ~57% at 3072 dimensions.
- Slope: Gentle upward trend.
2. **search-adaptor** (Orange crosses):
- Begins at ~51.5% at 128 dimensions.
- Sharp rise to ~57.5% at 256 dimensions.
- Continues upward to ~61% at 3072 dimensions.
- Slope: Steeper than Original(MRL), overtakes it at ~256 dimensions.
3. **MRL-Adaptor** (Green triangles):
- Starts at ~54.5% at 128 dimensions.
- Rises steadily to ~61.5% at 3072 dimensions.
- Slope: Moderate upward trend, consistently above Original(MRL) and search-adaptor until ~768 dimensions.
4. **SMEC** (Yellow squares):
- Begins at ~56.5% at 128 dimensions.
- Increases to ~61.5% at 3072 dimensions.
- Slope: Steady upward trend, maintaining the highest performance across all dimensions.
### Key Observations
- **Performance Hierarchy**: SMEC > MRL-Adaptor > search-adaptor > Original(MRL) at higher dimensions.
- **Convergence**: Differences between methods narrow at 3072 dimensions (e.g., MRL-Adaptor and SMEC differ by ~0.5%).
- **Divergence**: search-adaptor surpasses Original(MRL) by ~256 dimensions and maintains a lead.
### Interpretation
The data suggests that **SMEC** and **MRL-Adaptor** are superior to the baseline **Original(MRL)** and **search-adaptor** methods, particularly at larger dimensions. The search-adaptor shows rapid improvement but fails to surpass MRL-Adaptor. The convergence of lines at higher dimensions implies diminishing returns or similar scalability across methods. SMEC’s consistent lead highlights its robustness, while MRL-Adaptor’s steady growth indicates effective adaptation. The Original(MRL) baseline serves as a reference for incremental improvements in the other methods.
