Intro to Deep Learning Foundations
Free Oracle Cloud Infrastructure 2025 AI Foundations Associate practice — 6 questions on Intro to Deep Learning Foundations, with explanations. No sign-up. Full 12-question mixed test →
A data science team is building a model to forecast hourly electricity demand using the past 30 days of readings, where each hour's demand depends heavily on the pattern of previous hours. Which deep learning architecture family is most appropriate for this sequential dependency?
RNNs and related sequence models are designed to handle ordered, time-dependent data by maintaining a hidden state that captures information from prior steps, which fits time-series forecasting.
Why does increasing the depth (number of layers) in a neural network generally improve its ability to model complex data, compared to a shallow network with the same total number of neurons?
Depth allows the network to build hierarchical representations, where early layers learn simple features and later layers combine them into increasingly abstract concepts, which is a core reason deep learning outperforms shallow models on complex data.
An analyst has a small structured dataset of 500 rows with 10 numeric features and needs to predict a binary outcome. Model interpretability and fast training time are also required. Which approach is most appropriate?
Classical ML algorithms are well suited for small, structured datasets where interpretability and fast training matter, since deep learning typically requires large amounts of data to outperform simpler models.
In the transformer architecture that underlies modern large language models, what is the primary mechanism that allows the model to weigh the relevance of different words in a sequence relative to one another, regardless of their distance apart?
Self-attention is the core mechanism in transformers that allows every token to attend to every other token in the sequence simultaneously, capturing long-range dependencies without sequential processing.
A team is training a deep neural network and notices that during backpropagation, the gradients in the earliest layers become extremely small, causing those layers to stop learning effectively. What is this phenomenon commonly called, and what architectural element is often adjusted to help address it?
The vanishing gradient problem occurs when gradients shrink as they propagate backward through many layers, and using activation functions like ReLU (instead of sigmoid or tanh, which saturate) helps mitigate this by maintaining stronger gradient flow.
A company wants to build a system that classifies medical images (X-rays) into categories such as 'normal' or 'pneumonia detected.' Which deep learning architecture family is specifically designed to exploit the spatial structure of image data, such as local pixel patterns and shapes?
CNNs use convolutional filters that scan across an image to detect local spatial patterns such as edges, textures, and shapes, making them the architecture specifically designed for image data like X-rays.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →