liveBook · Manning

Deep learning

As explained in Hidden Influences

A machine learning approach that uses multilayered neural networks to learn hierarchical representations from data.

A social media feed may use clues such as when someone logs in to predict which posts they will engage with. Deep learning is a machine learning approach that uses layers of a neural network to learn increasingly abstract representations from data. section 2.7

Instead of manually specifying every useful higher-level feature, designers supply available data to a model, which discovers features and weights during training. For example, successive layers might use login times to develop a representation of someone who is more active in the morning. section 2.4 section 2.7

In a recommender system, a trained model can produce embeddings: internal numerical representations of users and items. These representations can help predict engagement, but their individual numbers generally cannot be decoded into human-readable preferences. section 2.7

Beyond a ratings table

A user-item ratings table records how users rated items but may miss context such as location or day of the week. Deep learning made data-driven recommendation attractive to online platforms as the technique became faster. Recommendations became more engaging, and users stayed longer on platforms that deployed them. section 2.3.2 section 2.4

The choice of what to optimize matters: it guides how a model is tuned. Greater engagement is one possible result, but deep learning also brings a transparency trade-off; giving people intelligible explanations of its decisions remains an open research problem. section 2.3.2 section 2.4

How Hidden Influences covers it

How deep it goes
Explains layered feature discovery and embeddings, while noting that human-readable explanations remain an open problem.
Examples
Uses login times and user-item ratings to illustrate learned representations and the limits of simpler recommendation data.
What you can do after
Evaluate and understand how learned representations support recommendations and why their decisions can be hard to explain.
How it connects
Connects neural networks to recommender systems, feature learning, embeddings, and optimization metrics.

Also explained in

Compare how 20 books explain deep learning →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Hidden Influences. The section links point to the parts of the book it draws on.