Natural Gradient
المُتدرِّج الطبيعي
تحسين لقاعدة التعلّم يضرب المُتدرِّج في مصفوفة موجبة التعريف (عادةً W^T W) فيُلغي الحاجة إلى عكس المصفوفة ويُسرّع التقارب بشكل ملحوظ. اقترحه أماري (1996).
An improvement to the learning rule that multiplies the gradient by a positive-definite matrix (typically W^T W), eliminating the need for matrix inversion and significantly speeding up convergence. Proposed by Amari (1996).
Also translated asالانحدار التدريجي الطبيعي
First appears in this corpus in: An Information-Maximization Approach to Blind Separation and Blind Deconvolution (1995)
Appears in these papers
- An Information-Maximization Approach to Blind Separation and Blind Deconvolution1995in the sky ✦
- Optimizing Neural Networks with Kronecker-Factored Approximate Curvature2015in the sky ✦
- Optimizing Neural Networks with Kronecker-Factored Approximate Curvature2015in the sky ✦
- Natural Gradient Works Efficiently in Learning1998in the sky ✦
- Trust Region Policy Optimization2015in the sky ✦
- Trust Region Policy Optimization2015in the sky ✦