Update README.md
Browse files
README.md
CHANGED
|
@@ -59,7 +59,7 @@ The test contains 256 fixed, geographically disjoint terrain–start pairs. Lear
|
|
| 59 |
|
| 60 |
| Method | Final coverage | Coverage–energy AUC | Coverage at 320 | Energy to 50% (censored; lower is better) |
|
| 61 |
|---|---:|---:|---:|---:|
|
| 62 |
-
| HUTACE | 0.7712 | 0.5213 | 0.5884 | 302.12 |
|
| 63 |
| ARiADNE-style Graph-Attention PPO (adapted) | **0.8553** | **0.5553** | **0.6166** | **272.35** |
|
| 64 |
| Plain U-Net PPO | 0.6679 | 0.4511 | 0.5090 | 369.77 |
|
| 65 |
| CNN PPO | 0.5454 | 0.4062 | 0.4725 | 408.95 |
|
|
@@ -71,10 +71,6 @@ HUTACE outperformed CNN PPO, Plain U-Net PPO, and all three fair partial-observa
|
|
| 71 |
|
| 72 |
The full-map greedy controller is a privileged diagnostic, not a fair baseline or formal optimality bound. Its descriptive results were 0.6567 final coverage, 0.4983 AUC, 0.5740 coverage at 320, and 321.99 censored energy to 50%.
|
| 73 |
|
| 74 |
-
## Ablation status
|
| 75 |
-
|
| 76 |
-
Phase 12 was skipped by explicit user direction. No component-level ablation was performed; this release makes no causal claim about individual HUTACE components.
|
| 77 |
-
|
| 78 |
## Efficiency
|
| 79 |
|
| 80 |
On a Tesla T4 with PyTorch 2.11.0, the representative seed-523 policy averaged 3.06 ms for GPU batch 1 and 0.291 ms per sample for batch 128. Single-thread CPU batch-1 latency on the same worker was 10.81 ms. The measured simulator CPU step averaged 1.09 ms. These are environment-specific measurements, not universal guarantees.
|
|
|
|
| 59 |
|
| 60 |
| Method | Final coverage | Coverage–energy AUC | Coverage at 320 | Energy to 50% (censored; lower is better) |
|
| 61 |
|---|---:|---:|---:|---:|
|
| 62 |
+
| HUTACE | **0.7712** | **0.5213** | **0.5884** | **302.12** |
|
| 63 |
| ARiADNE-style Graph-Attention PPO (adapted) | **0.8553** | **0.5553** | **0.6166** | **272.35** |
|
| 64 |
| Plain U-Net PPO | 0.6679 | 0.4511 | 0.5090 | 369.77 |
|
| 65 |
| CNN PPO | 0.5454 | 0.4062 | 0.4725 | 408.95 |
|
|
|
|
| 71 |
|
| 72 |
The full-map greedy controller is a privileged diagnostic, not a fair baseline or formal optimality bound. Its descriptive results were 0.6567 final coverage, 0.4983 AUC, 0.5740 coverage at 320, and 321.99 censored energy to 50%.
|
| 73 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 74 |
## Efficiency
|
| 75 |
|
| 76 |
On a Tesla T4 with PyTorch 2.11.0, the representative seed-523 policy averaged 3.06 ms for GPU batch 1 and 0.291 ms per sample for batch 128. Single-thread CPU batch-1 latency on the same worker was 10.81 ms. The measured simulator CPU step averaged 1.09 ms. These are environment-specific measurements, not universal guarantees.
|