Deep Learning Projects

Project 1 – Classification, weight sharing, auxiliary losses

This section details our approach to compare twodigits of a pair of two-channel images from the MNIST data.The goal of this project is to compare the accuracy of differentarchitectures and assess the performance improvement that canbe achieved through models with shared weights and the use of anauxiliary loss. The performance of each of these different modelswill be estimated on a test data through 10 rounds where bothdata and weight initialization are randomized at each trainin

• models.py contains the class definition of the different models we implemented.
• helpers.py contains a set of helper methods shared by the different models.
• test.ipynb is a Jupyter notebook, containing our code for performing hyperparameter optimization via grid search.
• test.py is a scrit which can be used to generate a random train and test dataset, and compare the performance of the three models, with different auxiliary loss weights. The models are pre-trained and are loaded from the pickles folder.

Project 2 – Mini deep-learning framework

The objective of this project is to design a mini “deep learning framework” using only pytorch’s tensor operations and the standard math library, hence in particular without using autograd or the neural-network modules.