Guide-yakabatana mibvunzo · Pakati Nhanho

Kuora Kwehuremu uye L2 Regularization Quiz

Edzai zvirango zveL2, kuderera kwakananga zero, SGD yakaenzana, AdamW decoupling, adaptive scaling, uye kuora simba mukudzidziswa.

Nzira dzekutungamira dzinoenderanaKuora Kwehuremu Uye L2 Regularization
Mubvunzo 1 ye 8

Chii chinonzi L2 kugadzika chinowedzera kune basa rekurasikirwa?

Tarisa mubvunzo wese8 mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

  1. Chii chinonzi L2 kugadzika chinowedzera kune basa rekurasikirwa?
    • Huwandu hwema layer mune network
    • Chirango pachiyero chekudzidza
    • Nhamba yemuenzaniso wekudzidzisa
    • Chirango chinoenzana nehuwandu hwehuremu hwakapetwa kaviri
  2. Ndeipi dambudziko guru iro kuora uremu kunobatsira kudzivirira?
    • Kutakura data zvishoma
    • Kuwedzeredza, uko modhi inorangarira ruzha mune data rekudzidzisa
    • Hardware kukundikana
    • Zvinyorwa zvisiri izvo
  3. Kuora uremu kunosundira kupi maremu emuenzaniso?
    • Kuenda kukusingaperi
    • Pakuremera kukuru
    • Kuenda ku zero
    • Kuenda kune zvakangoerekana zvaitika
  4. Nei AdamW yakaunzwa?
    • Kubvisa kuora kwehuremu kubva kune inogadzirisa gradient yekuvandudza kuitira kuti kuora kuite nemazvo
    • Kubvisa gradient zvachose
    • Kuita kuti kudzidzisa kunonoke pane chinangwa
    • Kubvisa kudiwa kwechiyero chekudzidza
  5. Kune akajeka stochastic gradient kudzika, L2 kugarisa uye kuora uremu kunobatana sei?
    • Vanogara vachipesana
    • Iwo akaenzana nemasvomhu
    • Kuora uremu kunodzima gradient
    • L2 inongoshanda kune SGD nekukurumidza
  6. Chii chinoitwa ne lambda coefficient control muL2 kugadzika?
    • Saizi yebatch
    • Nhamba yenguva
    • The activation function
    • Kusimba kwechirango pazviyero zvakakura
  7. Why can adding an L2 term to Adam’s loss differ from applying decoupled weight decay?
    • Adaptive per-parameter gradient scaling can distort the coupled penalty
    • Adam cannot train with any regularization
    • The two methods use different training labels
    • Decoupled decay removes the data loss
  8. In a simple decoupled update, what changes the fractional shrinkage of a weight?
    • Only the number of classes
    • Only the model’s input dimension
    • The learning rate and decay coefficient together
    • Only the validation-set size