Momentum Encoder
مُرمِّز الزخم
مُرمِّز لا يُدرَّب مباشرة بمتجهات الميل بل تُحدَّث أوزانه كمتوسط متحرك أُسِّي لأوزان مُرمِّز آخر (الاستعلام)، مما يضمن اتساق المخرجات عبر الزمن.
An encoder not trained directly by gradients but whose weights are an exponential moving average of another encoder (the query encoder), ensuring output consistency over time.
Also translated asالمُرمِّز الزخمي، المُرمِّز البطيء التحديث
First appears in this corpus in: Momentum Contrast for Unsupervised Visual Representation Learning (2020)
Appears in these papers
- Barlow Twins: Self-Supervised Learning via Redundancy Reduction2021in the sky ✦
- Contriever: Unsupervised Dense Information Retrieval with Contrastive Learning2022in the sky ✦
- Emerging Properties in Self-Supervised Vision Transformers2021in the sky ✦
- Momentum Contrast for Unsupervised Visual Representation Learning2020in the sky ✦
- Momentum Contrast for Unsupervised Visual Representation Learning2020in the sky ✦
- Exploring Simple Siamese Representation Learning2021in the sky ✦
- Exploring Simple Siamese Representation Learning2021in the sky ✦
- SwAV: Unsupervised Learning of Visual Features by Contrasting Cluster Assignments2020in the sky ✦