---

Due     April 2, 2025 by 11:59pm     Points     100     Submitting     a Google doc shared with Louisa (lconwill@nd.edu)

---

Take a pen and a paper, sketch the convolutional neural network specified below, and provide the total number of trainable parameters (weights and biases). You can put interim numbers on your sketch. Here’s the specification:

1. Input image:
• resolution: 11x11 pixels
• 3 channels (R,G,B)
2. Convolutional (“Conv2D”) layer:
• 3 kernels (feature maps)
• kernel size = (3,3)
• strides along spatial dimensions = (2,2)
• convolution type (padding) = 2
3. Fully-connected (“Dense”) layer:
• 2 neurons
4. Assume that all neurons in a single feature map have the same bias.

You can certainly code this first in e.g., Keras and call model.summary(). But you will lose the entire fun of thinking how the CNNs are built. You can also use ChatGPT as your "buddy". But be careful: for some unknown reason it gave me incorrect answer for this question multiple times, so use ChatGPT more creatively than asking to solve the quiz. Ideally, try to first solve it with a pen and a paper, and then code it in Python to verify your calculations. Submit your sketch as a solution. No need to submit the Python code.