KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Neural columns
Independent sub-networks added sequentially in the PNN architecture, specialized in learning new tasks without interfering with existing knowledge.
Lateral connections
Unidirectional links established from previous neural columns to new columns, enabling knowledge transfer without backpropagation to old knowledge.
Lifelong learning
Machine learning approach aiming to develop systems capable of learning continuously throughout their operational lifetime, accumulating and refining knowledge.
Neural architecture adaptation
Process of dynamically modifying the structure of a neural network in response to new learning requirements, typically by adding or removing components.
Column freezing
Technique consisting of freezing the weights of previous neural columns after their training to preserve knowledge and prevent degradation when adding new columns.
Lateral inhibition
Regulatory mechanism in PNNs where new columns can modulate the influence of previous columns to optimize the transfer of relevant knowledge.
Progressive expansion
Strategy of controlled increase in model capacity by sequentially adding computational resources only when necessary for new tasks.