Ikibazo gihuza ibibazo · Hagati Urwego

Kugabanuka k'uburemere na L2 Ikibazo

Gerageza L2 ibihano, kugabanuka kuri zeru, uburinganire bwa SGD, kugabanuka kwa AdamW, gupima imiterere, no kwangirika kwamahugurwa.

Inzira zijyanye no kuyoboraKugabanuka Ibiro na L2 Guhindura
Ikibazo 1 Bya 8

Niki L2 isanzwe yongeraho mubikorwa byo gutakaza?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Niki L2 isanzwe yongeraho mubikorwa byo gutakaza?
    • Umubare wibice murusobe
    • Igihano ku gipimo cyo kwiga
    • Kubara ingero zamahugurwa
    • Igihano gihwanye nigiteranyo cyibipimo bya kare
  2. Ni ikihe kibazo nyamukuru kugabanuka ibiro bifasha kwirinda?
    • Buhoro buhoro gupakira amakuru
    • Birenze, aho icyitegererezo gifata mu mutwe urusaku mumibare yamahugurwa
    • Kunanirwa kw'ibyuma
    • Ibirango bitari byo
  3. Ni kihe cyerekezo kubora gusunika uburemere bwikitegererezo?
    • Kugana ubuziraherezo
    • Kugana uburemere bunini
    • Kugana zeru
    • Kugana indangagaciro
  4. Kuki AdamW yatangijwe?
    • Kugabanya uburemere bwangirika uhereye kuri adaptive gradient update kugirango kubora bitware neza
    • Kuraho icyiciro cyose
    • Kugira ngo imyitozo itinde kubushake
    • Kurandura ibikenewe ku gipimo cyo kwiga
  5. Kubintu byoroshye bya stochastic gradient descente, L2 isanzwe hamwe no kugabanuka ibiro bifitanye isano ite?
    • Buri gihe barwanya
    • Bihwanye n'imibare
    • Kugabanuka k'uburemere bihagarika icyiciro
    • L2 ikora gusa kuri SGD ifite imbaraga
  6. Niki coefficient ya lambda igenzura muri L2 isanzwe?
    • Ingano yicyiciro
    • Umubare wibihe
    • Igikorwa cyo gukora
    • Imbaraga z'igihano ku buremere bunini
  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