Students should build a classifier for a multiclass image dataset such as CIFAR-10 or TinyImageNet.
Students should implement a convolutional neural network (CNN).
They should use convolutional and fully connected layers rather than attention mechanisms.
Students may start with any tutorial found online or in books, though they should build on that with original experimentation and observations.
See the slides from the Deep Learning lecture on Chapter 21. As before, the recommended platform is: Python, Jupyter Notebook, Tensorflow.
The grader will take into account correctness of the experiment, clarity of the code and explanation, rigor with which the model was evaluated, and cleverness in trying things.
Leave a Reply