Intro to Deep Learning Concepts and Applications (Day 1-Test 2)

What is the primary purpose of a Convolutional Neural Network (CNN) in image classification?

• The primary purpose of a CNN in image classification is to extract features and classify images.

• To convert images to grayscale

• To enhance image resolution

• To reduce the size of images

Describe the impact of overfitting on a CNN model's performance.

• Overfitting negatively impacts a CNN model's performance by reducing its ability to generalize to new, unseen data.

• Overfitting improves a CNN model's performance by enhancing its accuracy on training data.

• Overfitting allows a CNN model to better predict future data trends.

• Overfitting has no effect on a CNN model's performance or generalization capabilities.

What are the key steps involved in designing a CNN architecture for image classification?

• defining the problem, preprocessing data, choosing the architecture, selecting and training the model, evaluating the model, and fine-tuning.

• Ignoring data augmentation techniques

• Focusing solely on the output layer design

• Choosing the right optimizer only

How can you effectively collect and preprocess a dataset for training a CNN?

• Only collect data without cleaning

• Collect data, clean and label it, augment, normalize, and split into sets.

• Use unlabelled data for training

• Skip normalization and augmentation

What is the role of kernels in a CNN, and how do they affect feature extraction?

• Kernels in a CNN are used for feature extraction by convolving over the input data to detect specific features.

• Kernels in a CNN are responsible for pooling operations.

• Kernels are only used in the output layer of a CNN.

• Kernels are used to increase the size of the input data.

Explain the process of flattening in the context of CNNs and why it is necessary.

• Flattening is used to convert one-dimensional data into multi-dimensional arrays for CNNs.

• Flattening is the process of pooling data to reduce its size before passing it to the output layer.

• Flattening is the process of increasing the dimensions of data for better feature extraction.

• Flattening is the process of converting multi-dimensional data into a one-dimensional vector, necessary for input into fully connected layers in CNNs.

What techniques can be used to prevent underfitting in a deep learning model?

• Decrease model complexity

• Use fewer features

• Increase regularization

• Increase model complexity, use more features, reduce regularization, and train longer.

Describe the steps involved in training a CNN model and converting it for deployment.

• Model design, training, and skipping evaluation

• Data collection, model tuning, and ignoring deployment

• Data preprocessing, model evaluation, and deployment only

• The steps involved in training a CNN model include data collection, model design, compilation, training, evaluation, and tuning, followed by converting the model for deployment.

How do you run inference on a trained CNN model using new image data?

• Use the model to train on new image data.

• Run the model directly on the original image without any modifications.

• Load the model and run it without preprocessing the image.

• Load the model, preprocess the image, and run the model to get predictions.

What methods can be employed to evaluate and troubleshoot a CNN model's performance?

• Methods include loss/accuracy analysis, cross-validation, confusion matrices, examining misclassifications, and Grad-CAM.

• Adjusting learning rate only

• Implementing dropout layers

• Using random data generation