On a manufacturing production line producing children's toys, an inspection station is to classify different types of toys. An image classifier needs to be trained to perform the sorting. The following two datasets are available: DI: 100 labelled samples D2: 1 million labelled samples Provide brief answers for the following questions. (a) If dataset D1 is trained from scratch, state a problem that will occur. a . [1 marks] (b) Dataset D2 is trained from scratch, and then transfer learning is used to train on D2. Explain how this method solves the problem in part (a). [3 marks] (c) The model was trained and its training loss graph is shown in Figure 3(a). Explain this graph shape, then recommend a hyperparameter needs to be adjusted, and how must it be changed to improve the results? [3 marks) [] After making corrections, the model was trained again. The training and validation loss graph is shown in Figure 3(b). What hyperparameter needs to be adjusted, and how must it be changed to improve the results? (d) [3 marks) 1.0 1.0 Training loss 0.8 0.8 - Validation loss 0.6 9 0.4 Validation loss 0.6 9 0.4 Mu 0.2 0.2 0.0- 0 0.06 10 40 50 10 40 50 20 30 Epochs 20 30 Epochs (a) (b) Figure 3: Loss curves for Question 5 On a manufacturing production line producing children's toys, an inspection station is to classify different types of toys. An image classifier needs to be trained to perform the sorting. The following two datasets are available: DI: 100 labelled samples D2: 1 million labelled samples Provide brief answers for the following questions. (a) If dataset D1 is trained from scratch, state a problem that will occur. a . [1 marks] (b) Dataset D2 is trained from scratch, and then transfer learning is used to train on D2. Explain how this method solves the problem in part (a). [3 marks] (c) The model was trained and its training loss graph is shown in Figure 3(a). Explain this graph shape, then recommend a hyperparameter needs to be adjusted, and how must it be changed to improve the results? [3 marks) [] After making corrections, the model was trained again. The training and validation loss graph is shown in Figure 3(b). What hyperparameter needs to be adjusted, and how must it be changed to improve the results? (d) [3 marks) 1.0 1.0 Training loss 0.8 0.8 - Validation loss 0.6 9 0.4 Validation loss 0.6 9 0.4 Mu 0.2 0.2 0.0- 0 0.06 10 40 50 10 40 50 20 30 Epochs 20 30 Epochs (a) (b) Figure 3: Loss curves for Question 5