Exam AWS Certified Machine Learning - Specialty topic 1 question 29 discussion

The model must have been overfitted. Regularization helps to solve the overfitting problem in machine learning (as well as data augmentation). Correct answers should be BE.

Answer: BE https://www.kdnuggets.com/2019/12/5-techniques-prevent-overfitting-neural-networks.html 5 techniques to prevent overfitting: 1. Simplifying the model 2. Early stopping 3. Use data argumentation 4. Use regularization 5. Use dropouts

he issue described suggests that the model is overfitting to the training data: Training error decreases quickly, meaning the model is learning the training set very well. Poor performance on unseen test data, indicating overfitting. To resolve overfitting, the Machine Learning Specialist should: Perform Data Augmentation (B) Expands the training dataset artificially by applying transformations (e.g., rotations, flips, brightness changes, cropping). Helps the model generalize better by exposing it to more diverse variations of the same class. Add L2 Regularization (E) Also known as weight decay, it penalizes large weights, preventing the model from memorizing the training data. Encourages simpler models, which reduces variance and improves generalization.

agreed with vetal

A. NO - vanishing gradient is somebody bad they might happen and prevent convergence, we don't want that or something we can add explicitly. it is a result of the learning B. YES - we have a overfitting problem so more training examples will help C. NO - we already have good accuracy on the training set D. NO - gradient checking is to find bugs in model implementation E. YES - we have a overfitting problem

B. Perform data augmentation on the training data. ( it should add validation data as well) data should be distributed among train validation and test.

Answer B&E looks good

B & E is the correct ans

BE is exact

Looks like B and D are correct.. For D -> https://www.youtube.com/watch?v=P6EtCVrvYPU

L2 regularization tries to reduce the possibility of overfitting by keeping the values of the weights and biases small.

why not becuase of vanishing gradient?

This is L2 Regularization....Do you think this is the right answer?

agree BE