Catastrophic Forgetting

النسيان الكارثي

ميل الشبكات العصبية لفقدان المعرفة المكتسبة سابقاً بشكل مفاجئ حين تُضبط دقيقاً على بيانات جديدة، لأن مُتجهات الميل الجديدة تُعيد كتابة الأوزان التي ترمِّز المعرفة القديمة.

The tendency of neural networks to abruptly lose previously learned knowledge when fine-tuned on new data, because the new gradients overwrite weights encoding old knowledge.

Also translated asالفقد المعرفي المفاجئ للشبكة، الفقدان الكارثي، المحو التلقائي للمعرفة السابقة عند التحديث، النسيان التداخلي الكارثي

First appears in this corpus in: Human-Level Control Through Deep Reinforcement Learning (2015)

Appears in these papers

Parameter-Efficient Transfer Learning for NLP2019 in the sky ✦
BLIP-2: Bootstrapping Language-Image Pre-Training with Frozen Image Encoders and Large Language Models2023 in the sky ✦
Human-Level Control Through Deep Reinforcement Learning2015 in the sky ✦
Flamingo: a Visual Language Model for Few-Shot Learning2022 in the sky ✦
Improving Language Understanding by Generative Pre-Training2018 in the sky ✦
Open X-Embodiment: Robotic Learning Datasets and RT-X Models2024 in the sky ✦
RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control2023 in the sky ✦
Universal Language Model Fine-Tuning for Text Classification2018 in the sky ✦
Universal Language Model Fine-Tuning for Text Classification2018 in the sky ✦
Voyager: An Open-Ended Embodied Agent with Large Language Models2023 in the sky ✦
Voyager: An Open-Ended Embodied Agent with Large Language Models2023 in the sky ✦