Data representation

As explained in Machine Learning System Design

A learned or constructed encoding of data that captures information in a form suitable for input to a machine learning model.

Data representation is an encoding of data used as input to an ML model. It may be manually engineered or produced by a more complex model; learned vectors are often called representations or embeddings. section 6.2.3

Use representations to make data useful across the ML lifecycle:

• Generate representations for users, goods, images, videos, text, or audio with a shared model, then let applied teams use them for specific business problems. section 6.2.3
• Reuse representations from pretrained models as inputs to a simple model when data is limited and the target task is relatively simple. section 8.4
• Specify how representations are generated when documenting ETL, filtering, and feature-engineering decisions. section 6.7
• Store embeddings centrally when doing so reduces repeated processing, keeps features consistent, or supports similarity search. section 11.4.1

Image classification baseline

Take the representation from the last backbone layer of an ImageNet-trained model and train a simple linear classifier for custom image classes. This can outperform more elaborate models when the dataset is small. section 8.4

Figure 11.12 A landscape of ML problems with various needs for a feature store

Use the landscape to compare where centralized storage helps, especially when embeddings join tabular, image, text, or review data in multimodal systems. section 11.4.1

How Machine Learning System Design covers it

How deep it goes

Defines representations as model inputs and gives reuse and storage patterns, while deferring detailed feature-engineering treatment.

Examples

Uses ImageNet image representations, linear classifiers, CLIP, SentenceTransformers, Qdrant, and Faiss as concrete artifacts.

What you can do after

Apply it via pretrained models to build shallow baselines, centralize embeddings, and retrieve similar items for downstream models.

How it connects

Connects feature engineering, embeddings, pretrained models, transfer learning, feature stores, and vector databases.

liveBookmanning.com

Read Machine Learning System Design on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Deep Learning with R, Third EditionA data representation is a particular encoding or view of data that can make a task easier to solve.
• 

  liveBookmanning.com
  Deep Learning with PyTorch, Second EditionAn internal data encoding produced or consumed by a model.
• 

  liveBookmanning.com
  Modern CA representation is the concrete encoding used to express a value inside a program or in an external file.

Compare how 4 books explain data representation →

Related concepts

• neural network layerproduced by · A neural network layer is a computational stage or component that transforms input representations into outputs within a neural network.
• schemacontrasts with · A data schema is a formal specification of the structure, fields, types, relationships, and constraints of data.
• language modelproduced by · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• neural networkproduced by · A neural network is a computational model of interconnected, parameterized units that learns transformations from inputs to outputs.
• valuecontrasts with · A value is the content, quantity, information, or worth attributed to an entity, property, expression, or outcome.
• tokenimplemented by · A token is a discrete unit of text, such as a word, word fragment, character, or symbol, represented and processed as an element by a language model.

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