All translations
Enter a message name below to show all available translations.
Found 3 translations.
| Name | Current message text |
|---|---|
| h English (en) | {| class="wikitable" |- ! Problem !! Symptom !! Common mitigations |- | '''Vanishing gradients''' || Early layers learn extremely slowly || {{Term|relu}} {{Term|activation function|activations}}, {{Term|residual connection|residual connections}}, {{Term|batch normalization|batch normalisation}}, careful initialisation |- | '''Exploding gradients''' || Loss diverges or produces NaN values || {{Term|gradient clipping}}, weight {{Term|regularization|regularisation}}, lower {{Term|learning rate}} |} |
| h Spanish (es) | {| class="wikitable" |- ! Problema !! Síntoma !! Mitigaciones comunes |- | '''Gradientes que se desvanecen''' || Las capas iniciales aprenden extremadamente despacio || {{Term|activation function|activaciones}} {{Term|relu|ReLU}}, {{Term|residual connection|conexiones residuales}}, {{Term|batch normalization|normalización por lotes}}, inicialización cuidadosa |- | '''Gradientes que explotan''' || La pérdida diverge o produce valores NaN || {{Term|gradient clipping|recorte de gradientes}}, {{Term|regularization|regularización}} de pesos, {{Term|learning rate|tasa de aprendizaje}} más baja |} |
| h Chinese (zh) | {| class="wikitable" |- ! 问题 !! 症状 !! 常见缓解措施 |- | '''梯度消失''' || 早期层学习极其缓慢 || {{Term|relu|ReLU}} {{Term|activation function|激活函数}}、{{Term|residual connection|残差连接}}、{{Term|batch normalization|批归一化}}、谨慎的初始化 |- | '''梯度爆炸''' || 损失发散或产生 NaN 值 || {{Term|gradient clipping|梯度裁剪}}、权重{{Term|regularization|正则化}}、较低的{{Term|learning rate|学习率}} |} |