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.

• ✗

  Machine learning that requires an internet connection to function

  Why it's wrong here

  Deep learning models do not inherently require internet connectivity; the architecture consists of layers of computation that are executed locally once weights are learned. An internet connection may be useful for downloading pre-trained weights, accessing cloud training infrastructure, or serving predictions via an API, but these are operational conveniences rather than defining characteristics. The core definition remains neural-network depth and hierarchical representation learning, not network connectivity.

• ✓

  Machine learning using neural networks with many layers to learn hierarchical representations

  Why this is correct

  Deep learning is a machine learning approach that uses artificial neural networks with numerous hidden layers (a deep architecture) to automatically learn hierarchical representations of data. Lower layers learn simple features such as edges or phonemes, while deeper layers learn complex abstractions that capture semantics and structure. It is this multi-layer, end-to-end learning of feature hierarchies, not dataset size or other factors, that makes a model 'deep.'

• ✗

  A technique for training models on extremely large datasets only

  Why it's wrong here

  Although large datasets often help deep neural networks generalize and reduce overfitting, dataset size is not the defining criterion of deep learning. A model can be deep with relatively small data if transfer learning or data augmentation is used, while a large dataset can be used to train a shallow model such as logistic regression or a single-layer network. Deep learning is defined by the number of stacked trainable layers and the resulting hierarchy of learned features, not by the volume of training examples.

• ✗

  Machine learning that digs deeply into structured databases

  Why it's wrong here

  The word 'deep' in deep learning refers to the depth of a neural network — the number of successive transformation layers — not to how thoroughly a system interrogates structured databases. Running SQL-style queries or scanning tables has nothing to do with the backpropagation-driven training of multi-layer neural networks. Even when deep learning is applied to structured or tabular data, the 'depth' is still about stacked hidden layers, not data retrieval depth.

Go deeper

Related to this question