What is contrastive learning in the context of multimodal deep learning? Pick the 2 correct responses below.

• A. In a multimodal context, usually, contrastive learning increases the similarity of representations across modalities for the different objects and decreases the similarity of representations across modalities for same objects.
• B. Contrastive learning is a technique used to train deep learning models by comparing similar and dissimilar inputs and optimizing the model to maximize the similarity between representations of similar inputs and minimize the similarity between representations of dissimilar inputs.
• C. Contrastive learning is a technique used to manipulate and analyze multimodal data using Generative AI.
• D. In a multimodal context, usually, contrastive learning decreases the similarity of representations across modalities for the same objects and increases the similarity of representations across modalities for different objects.
• E. In a multimodal context, usually, contrastive learning increases the similarity of representations across modalities for the same objects and decreases the similarity of representations across modalities for different objects.

What does mixed-precision training refer to?

• A. Training a model using incomplete or missing information from different modalities.
• B. Training a model using diverse data types while addressing challenges related to missing or incomplete information.
• C. Training a model using multiple precision levels, such as using both single-precision and double- precision floating-point numbers.
• D. Training a model using different types of data, such as text, images, audio, time series, and geospatial information.

You are tasked with developing an image processing model using machine learning. You need to classify thousands of labeled images of cats and dogs. Which algorithm is commonly used for image classification?

• A. Convolutional Neural Networks (CNN)
• B. K-Means Clustering
• C. Linear Regression
• D. Decision Trees

What characteristic of autoencoders makes them suitable for anomaly detection?

• A. Their capacity to learn a compressed representation of the data.
• B. Their ability to classify images with high accuracy.
• C. Their capability to predict future outcomes based on past data.
• D. Their function in enhancing the quality of images.