Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Qulqulka Miisaanka iyo Kediska Habaynta L2

Tijaabi ganaaxyada L2, hoos udhaca eber, u dhigma SGD, AdamW kala goynta, miisaan la qabsiga, iyo quwadda suuska ee tababarka.

Wadooyinka haga ee la xidhiidhaQulqulka Miisaanka Iyo Joogteynta L2
Su'aal 1 ee 8

Muxuu joogtaynta L2 ku darayaa shaqada luminta?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Muxuu joogtaynta L2 ku darayaa shaqada luminta?
    • Tirada lakabyada shabakada
    • Ganaax ku saabsan heerka waxbarashada
    • Tirinta tusaalooyinka tababarka
    • Ciqaab u dhiganta wadarta miisaan labajibbaaran
  2. Waa maxay dhibka ugu weyn ee qudhunka miisaanku ka caawiyaa ka hortagga?
    • Soo dejinta xogta oo gaabis ah
    • Qalab xad dhaaf ah, halkaas oo moodelku xafido buuqa xogta tababarka
    • Qalabku guul daraystay
    • Calaamadaha khaldan
  3. Jihogee ayuu qudhunka miisaanku u riixaa miisaanka tusaalaha?
    • Dhanka aan dhamaadka lahayn
    • Dhanka miisaanka ugu weyn
    • Dhanka eber
    • Dhanka qiyamka random
  4. Waa maxay sababta AdamW loo soo bandhigay?
    • Si loo kala saaro suuska miisaanka ee cusboonaysiinta gradient-ka la qabsiga si uu qudhunku u dhaqmo si sax ah
    • Si aad gabi ahaanba meesha uga saarto jaan-goynta
    • Si loo yareeyo tababarka si ulakac ah
    • Si meesha looga saaro baahida heerka waxbarasho
  5. Marka la eego faraca hoose ee stochastic stochastic, sidee L2 joogtaynta iyo qudhunka miisaanka ula xidhiidhaan?
    • Had iyo jeer waa iska soo horjeeda
    • Waxay u dhigmaan xisaab ahaan
    • Burburka miisaanka wuxuu curyaamiyaa qulqulka
    • L2 waxay kaliya ugu shaqeysaa SGD oo leh dardar
  6. Waa maxay isku-darka lambda ku xakameynayo habaynta L2?
    • Cabbirka dufcadda
    • Tirada waayaha
    • Shaqada firfircoonida
    • Awoodda ciqaabta ee miisaanka waaweyn
  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