Loss Landscape
سطح الخسارة
تمثيل مجازي لدالة الخسارة بوصفها سطحاً متعدّد الأبعاد تتوزع عليه الأحواض (النهايات الصغرى) والتلال (الحواجز) ونقاط السرج. المُحسِّن يسعى لإيجاد أدنى نقطة على هذا السطح.
A metaphorical representation of the loss function as a multi-dimensional surface with basins (minima), ridges (barriers), and saddle points. The optimizer seeks the lowest point on this surface.
Also translated asمشهد الخسارة، تضاريس دالة الخسارة
First appears in this corpus in: SGDR: Stochastic Gradient Descent with Warm Restarts (2017)
Appears in these papers
- Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets2022in the sky ✦
- Symbolic Discovery of Optimization Algorithms2023in the sky ✦
- Visualizing the Loss Landscape of Neural Nets2018in the sky ✦
- Neural Tangent Kernel: Convergence and Generalization in Neural Networks2018in the sky ✦
- SGDR: Stochastic Gradient Descent with Warm Restarts2017in the sky ✦
- Shampoo: Preconditioned Stochastic Tensor Optimization2018in the sky ✦
- Sharpness-Aware Minimization for Efficiently Improving Generalization2021in the sky ✦
- Sharpness-Aware Minimization for Efficiently Improving Generalization2021in the sky ✦