Explain the role of transfer learning in reducing computational costs and time in model development.
Instruction: Provide insights into how transfer learning can make model development more efficient.
Context: This question tests the candidate's ability to apply transfer learning strategically to optimize resources, a key consideration in large-scale machine learning projects.
Updated
Official answer available
Read the opening below, then unlock the full answer and practical guidance.
Transfer can save target-development work by reusing representations instead of learning them from random initialization. Freezing a backbone can reduce backward-pass work; it still requires the forward pass, and a large pretrained model can be expensive to serve. Fine-tuning is not automatically cheaper than training a smaller task-specific model.
Distinguish...