| Available | Last Update | |
|---|---|---|
| Exams | Yes | 2026 |
| Homework | Yes | 2019 |
| Notes | Yes | 2023 |
| Slides | Yes | 2026 |
| Author | Last Update |
|---|---|
| Sveva Pepe | 2019 |
| Author | Last Update |
|---|---|
| Veronica Romano | 2021 |
| Author | Last Update |
|---|---|
| Sveva Pepe | 2020 |
| Author | Last Update |
|---|---|
| Alessandro Carotenuto | 2023 |
Project development
-
Solve the two classification problems: A) optimization prediction, B) compiler prediction.
For each classification problem, realize at least two variants (varying feature extraction, learning algorithm, learning hyper-parameters, etc.). Note: Use any method at your choice, except neural networks.
-
Evaluate each variant in a proper way. Find the best model and motivate the choice.
-
For each classification problem, apply the best model to predict output for the blind test set.
-
Write a report (about 10 pages) explaining all the work done: design and implementation choices, evaluation procedure and results. Reports must be individual.
Run the file with Google Colab, using the GPU. It's free.
Project development
-
Solve the image classification problem in two modes: A) define a CNN and train it from scratch, B) apply transfer learning and fine tuning from a pre-trained model. For training you can use any subset of the MWI dataset.
-
Evaluate the two models in a proper way. Discuss the best model and motivate the choice. Testing can be done either with cross-validation on MWI or with the SMART-I weather test set.
-
Write a report (about 10 pages) explaining all the work done: design and implementation choices, evaluation procedure and results. Reports must be individual.
-
Submit a set of images to be used to define a new test set
-
Submit your best model (layout and weights)
Run the file with Google Colab, using the GPU. It's free.
Here we have 2 files, because a training was done on a standard dataset and the other on an external one that was provided to us in a seminar.
Standard Dataset
External Dataset