Exponential Moving Average
المتوسط المتحرك الأُسّي
طريقة لحساب متوسط مُرجَّح لسلسلة زمنية يُعطي وزناً أكبر للقيم الحديثة ووزناً يتناقص أُسّياً للقيم القديمة. يُستخدم في RMSProp وAdam بدلاً من المجموع التراكمي في AdaGrad لمنع تآكل معدّل التعلّم.
A method for computing a weighted average of a time series that gives more weight to recent values and exponentially decreasing weight to older values. Used in RMSProp and Adam instead of AdaGrad's cumulative sum to prevent learning rate decay.
Also translated asالمعدّل المتحرك الأُسّي، EMA
First appears in this corpus in: Some Methods of Speeding Up the Convergence of Iteration Methods (1964)
Appears in these papers
- Adam: A Method for Stochastic Optimization2014in the sky ✦
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift2015in the sky ✦
- Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning2020in the sky ✦
- Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning2020in the sky ✦
- Consistency Models2023in the sky ✦
- Diffusion Policy: Visuomotor Policy Learning via Action Diffusion2023in the sky ✦
- DINOv2: Learning Robust Visual Features Without Supervision2023in the sky ✦
- Momentum Contrast for Unsupervised Visual Representation Learning2020in the sky ✦
- Momentum Contrast for Unsupervised Visual Representation Learning2020in the sky ✦
- Some Methods of Speeding Up the Convergence of Iteration Methods1964in the sky ✦
- Some Methods of Speeding Up the Convergence of Iteration Methods1964in the sky ✦
- RMSProp: Divide the Gradient by a Running Average of Its Recent Magnitude2012in the sky ✦
- VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning2021in the sky ✦