Machine learning uses algorithms to parse data, learn patterns or decision functions, and apply what it learns to make informed decisions. Deep learning is a specialized form that uses programmable neural networks. section 1.0
A production machine-learning workflow connects product requirements with deployed model services:
- Define the user problem and measurable requirements, such as intent recognition, prediction rate, and latency. section 1.1.3
- Explore available training data and consult researchers on algorithm choices. section 1.1.3
- Prototype training code, datasets, algorithms, and experimental models. section 1.1.3
- Evaluate alternatives and select a model for production. section 1.1.3
- Operationalize the model through continuous data processing, training, deployment, evaluation, and serving pipelines. section 1.1.3
Natural-language customer support
A customer-support application can replace a confusing multi-level menu with a question-based interface. A model recognizes the customer’s intent and matches the question with relevant answers. section 1.1.3
