Gradient Clipping Quiz

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. Ko clip-by-norm gradient clipping inonyanya kuchengetedza paunenge uchidzikamisa iyo update?
   • Iyo nzira yeiyo yakazara gradient yekuvandudza
   • Hukuru chaihwo hwegadhendi
   • Nhamba yemuenzaniso parameters
   • The kudzidza rate purogiramu
2. Mu clip-by-norm ine pachikumbaridzo c, chii chinoitika kana gradient yetsika g_norm ikadarika c?
   • Gradients akaiswa ku zero
   • Imwe neimwe gradient inowedzerwa ne c / g_norm
   • Kudzidzira kunomira
   • Chiyero chekudzidza chakapetwa kaviri
3. Nderipi dambudziko rakagadzirirwa gradient clipping kurwisa?
   • Vanishing gradients
   • Kutakura data zvishoma
   • Kuputika kwemagetsi
   • Overfitting kune yekusimbisa set
4. Ko clip-ne-value inosiyana sei kubva clip-by-norm?
   • Inogara ichichengetedza nzira yekuvandudza
   • Inoshanda chete pane convolutional layers
   • Inobvisa gradients zvachose
   • Inosungirira chikamu chega chega chegradient chakazvimirira kuita renji, chinogona kushandura gwara
5. A minibatch produces an unusually large global gradient norm. What does clip-by-norm do before the optimizer update?
   • Scales the full gradient vector down to the configured threshold when the norm exceeds it
   • Changes the model’s labels
   • Drops every gradient component independently
   • Increases the learning rate until the loss recovers
6. Mukusiyana kwakasiyana kudzidziswa (DP-SGD), ndeipi basa rinoitwa pa-emuenzaniso gradient clipping?
   • Iyo encrypts iyo data
   • Inosunga pesvedzero yemuenzaniso wega wega kuitira kuti ruzha rwakagadziriswa rugone kuvimbisa kuvanzika
   • Inowedzera batch size
   • Inobvisa kudiwa kwebasa rekurasikirwa
7. The global gradient norm is twice a clipping threshold c. By what factor is the gradient scaled?
   • Vaviri
   • One quarter
   • One half (c divided by 2c)
   • It is left unchanged
8. Why does DP-SGD clip each example’s gradient before adding calibrated noise?
   • To make all examples have the same label
   • To remove the need for a privacy budget
   • To increase the norm of rare-class gradients
   • To bound an individual example’s influence on the released update