Heat Equation Parameter-to-Observable Mapping (1000 Samples, Test)
2.576MSEProposed approach
Evaluation Results
| Method | Links | |
|---|---|---|
| Proposed approachDescription=Topological derivative approach where a new layer is added and retrained along with the first and last layers2025.02 | 2.576 | |
| Exhaustive searchSearch case=Best case, Description=Retraining the last layer, first layer and a randomly chosen hidden layer2025.02 | 2.604 | |
| Exhaustive searchSearch case=Worst case, Description=Retraining the last layer, first layer and a randomly chosen hidden layer2025.02 | 3.24 | |
| Complete retrainingDescription=Entire pre-trained network is trained to fit the new data2025.02 | 3.572 | |
| Traditional transfer learningDescription=Retraining only the last layer in the network2025.02 | 5.04 |