Olvido catastrófico is a phenomenon observed in inteligencia artificial, particularly in aprendizaje automático models, where a model forgets previously learned information upon learning nuevos datos or tasks. This issue is especially prominent in neural networks that are trained sequentially on different tasks, leading to a decline in performance on earlier tasks.
El término fue descrito por primera vez en el contexto de la psicológica teoría del aprendizaje, but it has significant implications in AI and machine learning. When a model is trained on one task and then retrained on a different task, it often fails to retain the knowledge acquired from the first task, resulting in poor performance on it. This is problematic for applications where continuous learning and adaptation are required.
Catastrophic forgetting is commonly attributed to the way neural networks update their weights during training. When new data is introduced, the adjustments made to the network’s parameters can interfere with the previously established representations, causing them to degrade or be overwritten. As a result, the model exhibits a high degree of sensitivity to the order in which tasks are presented.
Para mitigar el olvido catastrófico, los investigadores han explorado varias técnicas, incluyendo:
- Regularización métodos: These help maintain the weights associated with previous tasks while allowing new learning.
- Redes con memoria aumentada: These incorporate memoria externa estructuras para almacenar información de tareas anteriores.
- Redes neuronales progresivas: These architectures allocate separate resources for each task, preserving knowledge from earlier tasks.
Comprender y abordar el olvido catastrófico es crucial para desarrollar sistemas de IA that can learn continuously and adapt to new information without losing valuable knowledge.