AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Reinforcement learning

  Why this is correct

  In reinforcement learning, the robot learns by interacting with the maze environment and receiving a scalar reward signal for each action it takes. Through trial-and-error, it updates a policy to maximize cumulative discounted reward, rather than relying on pre-labeled examples. This direct feedback loop—action, reward, policy adjustment—is what makes it ideally suited for sequential navigation tasks. The robot's goal is to discover actions that yield the highest long-term return, even if immediate steps are suboptimal.

• ✗

  Supervised learning

  Why it's wrong here

  Supervised learning uses a fixed dataset of input-output pairs, where each input has a known correct label or target value. A model learns a mapping from inputs to outputs by minimizing a loss function comparing predictions to ground-truth labels. In the maze task, there is no collection of pre-labeled state-to-action pairs; the robot must explore and receive rewards dynamically. Without an explicit target output for every state, supervised learning cannot provide the environmental feedback necessary for iterative adaptation.

• ✗

  Unsupervised learning

  Why it's wrong here

  Unsupervised learning seeks to discover hidden patterns or cluster structure in data without any label or reward feedback. Techniques like k-means clustering or principal component analysis identify groups or latent factors, but they do not assign credit for good or bad decisions. Training a robot to navigate a maze requires an evaluative signal to distinguish successful from unsuccessful moves, which unsupervised learning inherently lacks. Thus, while it might find useful representations of maze layouts, it cannot alone improve navigation through reinforcement.

• ✗

  Semi-supervised learning

  Why it's wrong here

  Semi-supervised learning is a hybrid approach that leverages a small set of labeled examples alongside a large pool of unlabeled data, typically to improve model accuracy in classification or regression. It still relies on a loss function derived from labels and does not incorporate a reward signal from an environment. For the maze navigation problem, no such labeled dataset exists, and the robot must infer good actions based on delayed rewards. The learning paradigm is fundamentally predictive rather than interactive, so it cannot optimize a sequential policy through trial-and-error.

Go deeper

Related to this question