Case context: Your team is building a computer vision system to detect cats in pictures, but the model's accuracy is not yet good enough. The team is under high pressure and brainstorming ideas to improve it. Some members suggest gathering more of the same cat photos, while others want to modify the neural network itself.

Question: Based on the team's brainstormed options, what are the primary network-related and data-related changes the team can consider to improve the cat detector, and what specific examples of dataset diversity could they look for?

Sample answer: The team can consider data-related changes: collecting more cat pictures and collecting a more diverse training set. Specifically, they can seek photos of cats in unusual positions, cats with unusual coloration, or pictures taken with a variety of camera settings. For network-related changes, they can try a bigger neural network (with more layers, hidden units, or parameters), try a smaller neural network, add regularization, or change the neural network architecture.

Key points:

• Distinguishes data-related improvements (collecting more pictures, increasing training set diversity) from network-related improvements (bigger/smaller network, regularization, changing architecture).
• Specifies examples of diverse training data: cats in unusual positions, unusual coloration, or shot with varied camera settings.
• Correctly references network adjustments such as training longer, trying bigger or smaller networks, or adding regularization.

Rubric: The answer must distinguish between data-related options (more data, diverse data) and network-related options (size, architecture, regularization) from the source. It must also list specific examples of diversity (unusual positions, unusual coloration, camera settings) mentioned in the text.