continual learning

Gold definitionUpdated Apr 2, 2026

Definition

Continual learning enables models to acquire new knowledge sequentially without forgetting previously learned information. It addresses catastrophic forgetting, a key challenge in AI systems that must adapt and learn over time.

At a glance

Executive summary

Continual learning allows AI models to keep learning new things over time without forgetting what they already know. This is crucial for AI that needs to adapt in real-world situations, avoiding the need to retrain the entire model from scratch every time new information arrives.

TL;DR

Continual learning teaches AI models to learn new skills sequentially without forgetting old ones, like how humans learn throughout life.

Key points

Enables AI models to learn incrementally from continuous data streams without losing prior knowledge.
Solves the problem of catastrophic forgetting, where new learning erases old knowledge.
Used by researchers in AI, robotics, NLP, computer vision, and developers of adaptive AI systems.
Contrasts with traditional batch learning, which requires retraining the entire model on all data (old and new) for every update.
Current research trends focus on efficient memory management, prompt-based adaptation, and robust knowledge transfer in dynamic environments.

Use cases

Robotics: A robot learning new manipulation tasks or navigating new environments without forgetting previous skills.
Personalized AI assistants: An assistant continuously learning user preferences and new facts without losing past knowledge.
Autonomous driving: Vehicles adapting to new road conditions, traffic rules, or object types encountered after initial training.
Medical diagnosis systems: Models incorporating new disease patterns or treatment protocols as they emerge, without forgetting established knowledge.
Financial fraud detection: Systems continuously updating to new fraud schemes while retaining detection capabilities for older patterns.

Also known as

Lifelong learning, incremental learning, sequential learning, online learning