Case context: A team supplements mobile-app training images with internet images. The internet images have much higher resolution and are framed differently from the mobile-app images. The network has to learn these additional properties.

Question: Using the source's reasoning, diagnose the capacity-related risk and state what distribution the team should protect capacity for.

Sample answer: The risk is that the network will spend part of its representational capacity learning the internet images' higher resolution and different framing. Because these properties differ greatly from those of mobile-app images, less capacity may remain for recognizing the mobile-app distribution. The team should protect capacity for mobile-app images because that is the distribution it cares about.

Key points:

• Higher resolution and different framing are source-specific properties.
• Learning them uses representational capacity.
• Less capacity may remain for the mobile-app distribution.
• The possible result is weaker target performance.

Rubric: The response should identify the internet-specific properties, describe their consumption of representational capacity, connect that consumption to possible harm on mobile-app recognition, and name mobile-app images as the distribution of interest.