CompTIA DataX DY0-001 (V1) Practice Question

The correct answer is Zero-shot learning (ZSL). ZSL is a machine learning paradigm specifically designed for scenarios where a model must classify data from classes not seen during training. It works by learning a mapping between the input data (e.g., content) and a shared semantic space that is defined by auxiliary information, such as textual descriptions or attributes. This allows the model to classify a new, unseen item by finding the semantic class description that is most similar in that shared space.

• Few-shot learning is incorrect because it requires a small number of labeled examples for each new class, whereas the scenario specifies that zero examples are available.
• Autoencoder-based anomaly detection is incorrect because while it can identify that a piece of content is novel or anomalous (by checking for high reconstruction error), it cannot intrinsically classify that content into a specific, named category using its semantic description. Its primary function is detection, not classification of new categories.
• Self-supervised learning is a representation learning technique where a model learns from the data itself by solving a pretext task, such as predicting a hidden part of the input. While the resulting representations are powerful, this method does not directly address the task of classifying unseen categories using external semantic descriptions.