After an exhaustive grid search, the optimal hyperparameters produced the following:✑ 68% accuracy on the training set✑ 67% accuracy on the validation setWhat can the machine learning specialist do to improve the system’s accuracy?
Upload the model to an Amazon SageMaker notebook instance and use the Amazon SageMaker HPO feature to optimize the model’s hyperparameters.
Add more data to the training set and retrain the model using transfer learning to reduce the bias.
Use a neural network model with more layers that are pretrained on ImageNet and apply transfer learning to increase the variance.
Train a new model using the current neural network architecture.
Explanations:
Using Amazon SageMaker HPO (Hyperparameter Optimization) could potentially improve the model, but the low accuracy is more likely due to insufficient training data rather than hyperparameter settings alone. HPO may not solve the fundamental issue of model bias.
Adding more data to the training set is a common approach to reduce bias and improve the model’s generalization. Since the accuracy is relatively low, the model might benefit from a more diverse set of images, especially with only 150 images per type.
Using a more complex neural network with more layers could increase variance and potentially lead to overfitting. The issue seems to be more related to bias rather than variance, making this approach unsuitable.
Training a new model with the current architecture is unlikely to solve the problem. The issue seems to be related to insufficient data and bias, so retraining without addressing this may not yield improved results.
I outline that the answer is:
Add more data to the training set and retrain the model using transfer learning to reduce the bias.
I plot that the answer is:
Add more data to the training set and retrain the model using transfer learning to reduce the bias.