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

The video tutorial discusses overfitting in deep learning models, explaining why it's important to test if a model can overfit on training data. It covers the process of overfitting using Resnet 18, observing loss reduction and accuracy improvements. The tutorial also addresses common issues like activation functions, learning rates, and optimizers, providing troubleshooting tips. Advanced solutions such as changing initialization techniques and network architecture are explored for persistent problems.

What is a key indicator that a deep learning model is functioning correctly?

• The model's loss decreases consistently.

• The model's loss increases over time.

• The model's accuracy decreases consistently.

• The model's accuracy remains constant.

During the overfitting process, which optimizer is used in the example?

• Adam

• RMSprop

• Cross-entropy

• SGD

If a model's accuracy does not improve, which of the following could be a potential issue?

• The model is using too many epochs.

• The activation function is not suitable.

• The optimizer is too fast.

• The dataset is too small.

What might you consider adjusting if the learning rate is not optimized?

• Switch to a different dataset.

• Modify the learning rate value.

• Change the activation function.

• Increase the number of layers.

Which optimizer is recommended for reaching the global minima?

• SGD

• Adam

• RMSprop

• Adagrad

What could be a reason for receiving NaN as a loss output?

• The model is overfitting.

• The dataset is too large.

• The learning rate is too low.

• The initialization is incorrect.

If all other solutions fail, what might be the last resort to solve learning issues?

• Add more data to the training set.

• Change the network architecture.

• Increase the number of epochs.

• Use a different optimizer.