Glossary term
Glossary term
Training and Fine-Tuning
Training a model on sequential tasks or data streams without catastrophically forgetting previously learned knowledge.
ServiceNow's AI models use continual learning to incorporate new product knowledge into deployed agents without full retraining - new documentation is incrementally fine-tuned while preserving existing capabilities.
EfficientFormer uses continual learning for on-device model updates on smartphones - when Apple releases a new iOS feature, relevant ML models are updated via federated continual learning without erasing prior skills.
PaLM 2 uses continual pre-training on a curated medical corpus to create Med-PaLM 2 - retaining general language skills while adding clinical knowledge through a staged continual training approach.