Final project of KTH course DD2424 Deep Learning in Data Science. This project consists of a simplified replication of the paper Attention Augmented Convolutional Networks. The main goal was augmenting Convolutional Networks by means of self-attention. As a result, we have convolutions that also detect long range interactions.
To run the experiments:
- Go to scripts.
- Select the architecture of the network.
- Modify the network parameters to customize the architecture.
- Run the selected code (it needs TensorFlow in its version 2 to run).
The conclsusions of the different experiments can be found in Group76 Project Report.