Repository for hosting deliverables in the course IT3105 Artificial Intelligence Programming at the Norwegian University of Science and Technology, fall 2015. Entire course completed as a group effort with Sondre Dyvik.
module1: Using A* to solve Navigation Problems.
module2: A*-GAC, A General Constraint-Satisfaction Problem Solver.
module3: Using A*-GAC to Solve Nonograms.
Project 2 - Implemented in Java using the course-provided GUI
module4: Minimax for Playing the 2048 Game.
Out of ten runs per heuristic, we achieved the following results:
| Heuristic | 1024 tile | 2048 tile | 4096 tile | Average score | Best score |
|---|---|---|---|---|---|
| Snake | 100% | 100% | 70% | 55158 | 79472 |
| Gradient | 100% | 90% | 50% | 48206 | 76756 |
| See the report in the project 2 directory for more details. |
Project 3 - Implemented in Python 3.4 using Theano
module5: Neural Networks for Image Classification
Five different neural networks were tested on the MNIST dataset, and we achieved the following results:
| Neural net | Minibatch size | Epochs | Learning rate | Training set average | Test set average | Average time trained |
|---|---|---|---|---|---|---|
| Net #1 | 10 | 30 | 10-2 | 99.16% | 97.71% | 263.57 s |
| Net #2 | 100 | 25 | 10-2 | 99.74% | 97.95% | 131.48 s |
| Net #3 | 15 | 30 | 5*10-2 | 99.73% | 97.64% | 95.91 s |
| Net #4 | 20 | 30 | 10-2 | 99.81% | 98.10% | 463.20 s |
| Net #5 | 15 | 20 | 10-2 | 98.31% | 97.34% | 244.96 s |
| See the module 5 report in the project 3 directory for more details. |
module6: Neural Networks for Game Playing
See the LICENCE file for license rights and limitations (MIT).