Distribution Shift
انزياح التوزيع
تغيّر بين التوزيع الإحصائي لبيانات التدريب والبيانات التي يواجهها النموذج عند النشر، وهو السبب الرئيسي لهشاشة النماذج المضبوطة دقيقاً.
A change between the statistical distribution of training data and the data encountered at deployment — the main cause of brittleness in fine-tuned models.
Also translated asالانحراف التوزيعي، الانزياح التوزيعي، تحوّل التوزيع، تغيُّر التوزيع
First appears in this corpus in: Conservative Q-Learning for Offline Reinforcement Learning (2020)
Appears in these papers
- Learning Transferable Visual Models from Natural Language Supervision2021in the sky ✦
- Learning Transferable Visual Models from Natural Language Supervision2021in the sky ✦
- Conservative Q-Learning for Offline Reinforcement Learning2020in the sky ✦
- Conservative Q-Learning for Offline Reinforcement Learning2020in the sky ✦
- Are Transformers Effective for Time Series Forecasting?2023in the sky ✦
- A Time Series Is Worth 64 Words: Long-Term Forecasting with Transformers2023in the sky ✦
- SWE-bench: Can Language Models Resolve Real-World GitHub Issues?2024in the sky ✦
- Robust Speech Recognition via Large-Scale Weak Supervision2022in the sky ✦