A repository containing my submissions for the evaluation test for prospective GSoC applicants for the DeepLense project. I applied for the Domain Adaptation for Decoding Dark Matter with Strong Gravitational Lensing project, and had to do the following tests:
- Common test (I): Simulating Superfluid Dark Matter (DM) with PyAutoLens: solution can be found here (
superfluid_dark_matter/simulation.ipynb)! - Specific test (II): Learning the DM representation: solution can be found here (
aae.ipynb)! NOTE: I had to submit the evaluation test until March 24 but I had a lot of university work at that period, as a result my first submission was pretty bad, I made a lot of mistakes and had a very underwhelming performance. The solution in the last link is the one where I fixed all of those mistakes and basically reworked everything that I did. For the sake of transparency though, my original submission can be found here (domain_adaptation/domain_adaptation.ipynb).
Below are the descriptions of each test:
Familiarize yourself with PyAutoLens: Refer to the following documentation for examples of how to simulate strong gravitational lensing.
What is PyAutoLens? — PyAutoLens
Task: Modify and/or use the already existing functionality of PyAutoLens to simulate strong lensing from superfluid dark matter. Specifically, you will need to simulate the effects of lensing from a linear mass density - imagine this being a string of mass on galactic scales.
Unsupervised Anomaly Detection and Transfer Learning
The following link contains the network architecture and weights of an Adversarial Autoencoder trained on a large dataset of strong gravitational lensing images without substructure.
Link: https://github.com/ML4SCI/ML4SCI_GSoC/tree/main/DeepLense/Domain%20Adaptation
Task: Use the provided model weights for weight initialization or feature extraction, and train an unsupervised deep learning algorithm of your choice to learn the distribution of the provided strong lensing images with no substructure to solve the task of unsupervised anomaly detection using PyTorch. Pick the most appropriate approach and discuss your strategy.
Dataset: https://github.com/ML4SCI/ML4SCI_GSoC/blob/main/DeepLense/Domain%20Adaptation/lenses.tgz
Dataset Description: A set of simulated strong gravitational lensing images with and without substructure.
Evaluation Metrics: ROC curve (Receiver Operating Characteristic curve) and AUC score (Area Under the ROC Curve)