Question: In a concise analytical response, explain why the risk of merging training data sources, such as user-uploaded images and internet images, depends on the flexibility of the learning algorithm being used. Contrast earlier algorithms with modern flexible algorithms as described in the source.

Sample answer: Earlier learning algorithms, such as hand-designed computer vision features followed by a simple linear classifier, had limited capacity to reconcile differences between data sources, so merging user-uploaded and internet images carried a real risk of worsening performance. Modern, powerful, flexible algorithms like large neural networks can better accommodate varied data, so this same merging carries greatly diminished risk.

Key points:

• Earlier algorithms (hand-designed features + linear classifier) had a real risk of worse performance from merging data
• Modern algorithms (large neural networks) have greatly diminished risk
• The determining factor is the flexibility/power of the algorithm
• The example concerns merging user-uploaded and internet images

Rubric: Full credit identifies both algorithm types (earlier hand-designed feature + linear classifier vs. modern large neural network), states the risk associated with each, and explains that algorithm flexibility is the underlying reason for the difference in risk.