Assignment 3: Encoder-decoder models (Deadline May 13!)

In this assignment, your task is to improve an existing encoder-decoder model. The assignment is described in this Colab notebook.

This assignment is very different from previous ones. You will be graded on the score that your model achieves on the test set, as calculated in the leaderboard. It's all explained in the Colab notebook.

Late policy: each late day (from the 1st minute overdue) reduces points by 10%. Decay is linear. E.g., if you submit a homework 1 day and 5 hours late (resulting in a reduction of 20%), and your real score is 11, your reduced score is 1*0.80=8.8≈9. That is, the number of maximum late days is 10.

Assignment must be completed by the student alone (consulting with co-students is of course allowed).