Adversarial Training for Multi-Task Learning
Note:
both share and private models are based on CNN, and the following table presents the average error rates on 16 datasets across different MTL architectures above.
| Type | (a) | (b) | (c) |
|---|---|---|---|
| Avg Err | 0.1585 | 0.1462 | 0.1312 |
Paper:
