Evaluate the impact of an AI application used in the real world. (case study) : Working with Flower Images: Case Study -

The video tutorial covers hyperparameter tuning, focusing on learning rates and optimizers. It introduces the concept of callbacks in Keras, explaining their role in managing training processes, such as saving models and stopping training early. The tutorial demonstrates creating custom callbacks for specific needs and discusses practical applications, including tuning learning rates with ResNet models. An interactive Q&A session explores various callback functions, encouraging further exploration and understanding.

What is a common hyperparameter that is often tuned in machine learning models?

• Batch size

• Learning rate

• Number of layers

• Activation function

What is the primary purpose of using callbacks in Keras?

• To reduce model size

• To increase model accuracy

• To perform actions during training

• To visualize data

Which of the following is an example of a custom callback action?

• Changing the optimizer

• Adjusting the batch size

• Modifying the dataset

• Saving training loss

What is the role of the 'history' object in the custom callback example?

• To store model weights

• To validate the model

• To adjust learning rates

• To track and save training loss

Why might a high learning rate be problematic in training?

• It can lead to overfitting

• It may cause the model to converge too slowly

• It can cause oscillations near the minimum

• It increases computational cost

What does the 'early stopping' callback do?

• Saves the model at regular intervals

• Increases batch size during training

• Stops training when a metric stops improving

• Reduces learning rate when loss plateaus

What is the function of the 'model checkpoint' callback?

• To terminate training early

• To increase model accuracy

• To save model weights after each epoch

• To adjust learning rates

How does the 'reduce LR on plateau' callback assist in training?

• By increasing the learning rate when loss decreases

• By reducing the learning rate when a metric plateaus

• By saving the model at the best epoch

• By stopping training when accuracy is high

What is a potential benefit of using custom callbacks?

• They reduce the need for data preprocessing

• They simplify the model architecture

• They increase the model's size

• They can automate repetitive tasks

Which callback would you use to terminate training if a non-numeric loss is encountered?

• Terminate on NaN

• Reduce LR on plateau

• Model checkpoint

• Early stopping