Learning algorithm

As explained in Designing Deep Learning Systems: A guide for software engineers

A procedure that fits a machine learning model to data by optimizing its parameters or objective.

A training algorithm is the procedure used to fit a machine learning model to data by optimizing its parameters or objective. It defines the model architecture and training plan, then runs as training code under a system that supplies CPU, GPU, memory, and—when needed—distributed computation. section 3.0 section 6.1.5

Select a training algorithm during research and prototyping by testing candidate approaches against available data and the product requirement. Accuracy alone is insufficient: compute needs, data volume, and implementation cost affect whether an approach is practical enough to adopt. section 1.1.1 section 10.1

Treat algorithm choice as part of an iterative production workflow:

• Explore data and identify approaches that can address the business requirement. section 1.1.1
• Build proof-of-concept models with multiple candidate algorithms and evaluate the resulting models. section 1.1.1
• Vet the selected data and algorithm with stakeholders before investing in production pipelines. section 1.1.1
• Stabilize training code and data shape, then onboard the chosen prototype into continuous training, deployment, evaluation, and serving pipelines. section 1.1.3 section 10.1

Customer-support intent matching

For a service that recognizes a customer question’s intent and matches relevant answers, prototype several deep learning models from collected training data. After evaluation, select one model, productionize its training code, and connect the deployed prediction service to the application backend under defined throughput and latency requirements. section 1.1.3

The same machine learning algorithm can appear in both training and serving, but it executes in different modes. Training produces model files; deployment copies those trained files into the production environment, where a prediction service loads them to handle requests. section 6.1.5

Figure 1.2 MLOps applies DevOps approaches to deep learning for the productionization phase, when models get shipped to production. (Source: Machine Learning Engineering in Action, by Ben Wilson, Manning, 2022, figure 2.7)

Use the productionization view to place training-algorithm work before MLOps practices that ship a model into production. section 1.1.1

For large datasets or model architectures, a training service becomes essential because it manages distributed execution and costly compute resources. Containerized training code can be launched and monitored through framework-specific Kubeflow operators, while Kubernetes APIs submit jobs and retrieve status or results. section 3.0 section 3.4.1

How Designing Deep Learning Systems: A guide for software engineers covers it

How deep it goes

Explains selection criteria, execution context, and production integration rather than algorithmic optimization internals.

Examples

Uses a customer-support intent-matching service that progresses from experimental models to a deployed prediction service.

What it assumes

Assumes no deep learning theory or framework knowledge, teaching only the training-code preparation needed for service work.

What you can do after

Evaluate/understand candidate algorithms by balancing model results with compute, data-volume, and implementation-cost constraints.

How it connects

Connects training algorithms with data exploration, model training services, model serving, deployment, and MLOps.

liveBookmanning.com

Read Designing Deep Learning Systems: A guide for software engineers on liveBook Get the book at manning.com

Related concepts

• software artifactimplemented by · A software artifact is a persistent work product or object produced or maintained during software development or a related workflow.
• language modelproduces · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.
• gradient descenthas part · Gradient descent is an iterative optimization method that updates parameters opposite the gradient of an objective function to reduce its value.
• objective functionhas part · An objective function assigns values to candidate solutions, predictions, or parameter settings according to a criterion to be minimized or maximized.
• model training serviceused by · A service that provides computation resources and job management for executing model-training code, producing trained models, and tracking training execution.
• model developmentused by · Model development is the process of designing, constructing, training, evaluating, and refining a computational or statistical model for a defined task.

This explainer was generated by Manning's knowledge graph from the text of Designing Deep Learning Systems: A guide for software engineers. The section links point to the parts of the book it draws on.