Ikibazo gihuza ibibazo · Biragoye Urwego

Icyiciro cya kabiri-Optimisation hamwe na Newton Methods Ikibazo

Gerageza uburyo kugabanuka bihindura intambwe yo gutezimbere, ni ubuhe buryo bwa Quasi-Newton bubika, kandi iyo imyitwarire ya Newton iterwa nintego ya quadratic.

Ikibazo 1 Bya 8

Ni ayahe makuru uburyo Newton akoresha ubwo buryo bwa gradient descente budakoreshwa?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Ni ayahe makuru uburyo Newton akoresha ubwo buryo bwa gradient descente budakoreshwa?
    • Kugabanuka, ukoresheje inkomoko ya kabiri muri matrike ya Hessian
    • Ahantu nyaburanga byibuze kwisi
    • Gahunda yo kwiga
    • Umubare w'ingero zamahugurwa
  2. Kugirango intego ya quadratic itunganijwe neza, ni inzira zingahe uburyo bwa Newton bukeneye kugera byibuze?
    • Intambwe icumi
    • Intambwe imwe
    • Umubare umwe nkuwamanutse buhoro buhoro
    • Ntishobora guhura
  3. Ni ukubera iki uburyo bwuzuye bwa Newton budashoboka kuri miliyari-ibipimo by'imitsi iva mu mitsi?
    • Ikigereranyo ntigishobora kubarwa
    • Ibikomokaho bya kabiri buri gihe ni zeru mumiyoboro ya neural
    • Hessian ni N-by-N, igura hafi ya N-kare yibuka na N-cubed compte kugirango ihindure
    • Ikora kumashusho gusa
  4. Ni ubuhe buryo bwa quasi-Newton nka BFGS bukora kugirango wirinde ikiguzi cya Hessian?
    • Birengagije icyiciro cyose
    • Bahitamo gukekeranya
    • Bahindura ikibazo kumurongo umwe
    • Bubaka ikigereranyo cya Hessian ihindagurika uhereye kubitandukanya bigenda bikurikirana
  5. Nigute L-BFGS igabanya kwibuka ugereranije na BFGS?
    • Irabika gusa icyiciro cya nyuma cya gradient nintambwe aho kuba matrix yuzuye
    • Ikoresha icyerekezo kimwe kireremba gusa
    • Ijugunya icyiciro nyuma ya buri ntambwe
    • Itoza ibipimo bike
  6. Nubuhe buryo bugezweho bugereranya kugabanuka kugirango byihute imyitozo nini ya nete?
    • Kode imwe
    • Uburyo bukoreshwa na Kronecker nka K-FAC na Shampoo
    • Ikibaya cya stochastic gradient kumanuka nta gihindutse
    • Kureka
  7. On an exact quadratic objective with an invertible Hessian, what assumption supports the one-step Newton result?
    • The learning rate is reduced to zero before the update
    • The Hessian is ignored and only a fixed scalar step is used
    • The full Newton update uses the exact gradient and Hessian of that quadratic
    • The objective is first converted to a linear function
  8. A non-quadratic objective is optimized with Newton’s method. What role does its quadratic model play?
    • It guarantees the global minimum in one step
    • It replaces the need to evaluate a gradient
    • It makes every Hessian positive definite
    • It approximates local curvature, so later iterations may be needed as the model changes