AI-ordlista
Den kompletta ordlistan över AI
Catastrophic Forgetting
Phenomenon where neural networks completely forget prior knowledge when learning new tasks.
Elastic Weight Consolidation
Regularization technique that identifies and protects important neural weights for previous tasks.
Experience Replay
Method that stores and reuses past examples to maintain performance on learned tasks.
Progressive Neural Networks
Architecture that allocates new neural columns for each task while maintaining connections to previous tasks.
Plasticity-Stability Dilemma
Fundamental challenge of balancing the ability to learn new information and the preservation of existing knowledge.
Meta-Learning
Approach where the model learns to learn, developing rapid adaptation strategies to new tasks.
Curriculum Learning
Training strategy presenting examples in increasing order of difficulty to facilitate continuous learning.
Knowledge Distillation
Technique transferring knowledge from an old model to a new one to preserve learned capabilities.
Memory-Based Continual Learning
Approaches using external memories to selectively store and retrieve information from past tasks.
Parameter Isolation
Methods allocating dedicated subnetworks for different tasks to avoid interferences.
Unsupervised Continual Learning
Unsupervised continuous learning, exploiting intrinsic data structures for knowledge acquisition.
Online Continual Learning
Scenario where data arrives in a continuous stream, requiring immediate adaptation without the possibility of revisiting past examples.