Exam AWS Certified Machine Learning - Specialty topic 1 question 157 discussion

B CE is correct.

(Option C): Using open source libraries to crop, resize, flip, rotate, and adjust the brightness and contrast of the images can increase the diversity of the training data, helping the model generalize better to new products1. (Option D): Normalizing and scaling the images can help the model learn more effectively by ensuring that the input data is consistent2. (Option F): Addressing class imbalances can prevent the model from being biased towards more frequent classes, improving its overall accuracy.

The questions says "The images for each product are classified according to specific product lines." why do we need Amazon Rekognition Custom Labels then?

Option C is correct because augmenting the images in the dataset can help the model learn more features and generalize better to new products. Image augmentation is a common technique to increase the diversity and size of the training data. Option E is correct because Amazon Rekognition Custom Labels can train a custom model to detect specific objects and scenes that are relevant to the business use case. It can also leverage the existing models from Amazon Rekognition that are trained on tens of millions of images across many categories. Option F is correct because class imbalance can affect the performance and accuracy of the model, as it can cause the model to be biased towards the majority class and ignore the minority class. Applying oversampling or undersampling can help balance the classes and improve the model's ability to learn from the data

assuming improve accuracy of the (existing) solution

Hopefully final answer this time CEF. was initially looking for D but changed to E now

C & F for sure the confusion between D and E but lets go for D as E will need more steps

The question asks for quick solutions and to improve the classifier's accuracy. Since we want a quick fix, I'm going to avoid solutions that requires a new model implementation. Therefore, the alternatives that can improve the performance of the current classification are: Letter F, C and D. Letters B and E would bring a new development cost from zero and Letter A does not solve the classification problem.

the three steps that would improve the accuracy of the solution are C (data augmentation), D (image normalization and scaling), and F (addressing class imbalances)

CEF : C&F for Overfitting; E : "Rekognition DetectLabel" is the general image labeling capability of Amazon Rekognition, which provides predefined labels for common objects and concepts out-of-the-box. On the other hand, "Rekognition Custom Labels" allows you to create custom models to detect specific labels or objects that are not covered by the default labels,

CEF better choose

This is CDF. No idea why this is unclear here.

The problem is about "Overfitting", because the new products doesn't work well. It is not about simply improve model accuracy. C is great answer, augmentation is for overfitting. D is wrong, because normalization of pixel is not for overfitting, and "all images have the same dimensions. no need for scaling, they are already scaled. F is for imbalance data. if the data is imbalanced, they should perform poor on both training and testing data(new product). And the new product should perform bad only on those cold category, not overall poor performance. B and E are all about Rokognition, one is Rekognition Detect label, a built-in image classification model; one is Rekognition Custom Labels, a pre-trained with fine-tuning model.

you fix images (C), train Rekognition with these images (E) and finally infer to get the classes (B)

Amazon Rekognition custom label model requires time, expertise, and resources, often taking months to complete. Additionally, it often requires thousands or tens-of-thousands of hand-labeled images to provide the model with enough data to accurately make decisions. The solution must be quick. https://aws.amazon.com/rekognition/custom-labels-features/

CEF is correct

D seems doing nothing with new product