This is my personal repository to submit my solutions to the test problems provided by ML4SCI as a selection test for GSOC'24.
Title | Layout | Project | Description |
---|---|---|---|
Multi-Class Classification (Common Test 1) | gsoc_proposal | DEEPLENSE | Implementing multi-class classification. |
DDPM (Specific Test 4) | gsoc_proposal | DEEPLENSE | Implementing denoising diffusion probabilistic model. [This will be my GSOC'24 proposal project] |
SSL on Real Dataset | gsoc_proposal | DEEPLENSE | Implementing SSL Contrastive learning |
Image generated at first epoch Image generated at 20th epoch
Model Architecture | Number of epochs | Learning rate | Training Loss | Training Accuracy | Micro-average ROC-AUC | Macro-average ROC-AUC |
---|---|---|---|---|---|---|
ResNet18 | 50 | 1e-4 | 0.00114 | 1 | 1.0 | 1.0 |
CNN | 50 | 1e-4 | 1.03 | 0.461 | 0.6626 | 0.64877 |
Model Architecture | Number of epochs | Learning rate | Training Loss | Training Accuracy | Lensed AUC | Non Lensed ROC-AUC |
---|---|---|---|---|---|---|
Supervised Learning | 20 | 1e-4 | 0.7248 | 0.7083 | 0.9 | 0.9 |
SSL with Rotation Augmentation | 20 | 1e-4 | 0.7336 | 0.6369 | 0.82 | 0.73 |
SSL with Gaussian Augmentation | 20 | 1e-4 | 0.6209 | 0.7619 | 0.91 | 0.91 |
Implemented a DDIM as a part of the Specific Test 4. 2 images generated from that model are shown below: