AWS Certified AI Practitioner Practice Exam

What ML strategy allows a company to adapt pre-trained models for new related tasks without starting from scratch?

1. Increase the number of epochs

2. Use transfer learning

3. Decrease the number of epochs

4. Use unsupervised learning

The correct answer is: Use transfer learning

The concept of transfer learning is pivotal in the field of machine learning, especially for organizations looking to leverage existing models for new, but related tasks. Transfer learning enables a company to adapt a pre-trained model that has already learned a significant amount about a particular domain, allowing it to be fine-tuned or modified for a similar task. This process saves time and computational resources since it eliminates the need to start from scratch with training a model on a large dataset. By using transfer learning, a company can take advantage of the knowledge that the pre-trained model has already acquired, which can be beneficial when data for the new task is limited or costly to obtain. The adaptations may involve retraining certain layers of the model while keeping others frozen, thereby allowing the model to maintain its learned general characteristics while becoming specialized for the new task. In contrast, increasing or decreasing the number of epochs relates to how long a model is trained, but does not inherently involve using an existing model for a new application. Unsupervised learning, on the other hand, focuses on identifying patterns in data without labeled outputs and is not specifically designed for adapting pre-trained models to new tasks. This clearly highlights why transfer learning is the optimal strategy in the context presented.