Training is where a model learns. An algorithm takes a dataset, often labeled, and repeatedly adjusts parameters until the model’s outputs match the expected answers well enough. For deep networks this is compute-intensive and usually done on servers with GPUs, far from the devices that will eventually run inference.
In IoT the hard part is rarely the algorithm but the data: sensor readings have to be collected, cleaned, labeled and stored, often as time-series data from many sites. Training also does not end at launch; as equipment ages or environments change, model drift sets in and periodic retraining is needed. Federated learning is one approach that trains on devices without centralizing raw data.
Part of the IoT Glossary. Spotted an error or a missing term? Tell us.