Maswali ya Uboreshaji wa Agizo la Pili na Maswali ya Mbinu za Newton

Majibu na maelezo huonekana unapojibu maswali.

1. Je, ni habari gani ambayo njia ya Newton hutumia ambayo asili ya upinde rangi isiyo na kipenyo haifanyi?
   • Mviringo, kupitia viasili vya pili kwenye tumbo la Hessian
   • Mahali halisi ya kima cha chini cha kimataifa
   • Ratiba ya kiwango cha kujifunza
   • Idadi ya mifano ya mafunzo
2. Kwa lengo la mara nne kamili, ni hatua ngapi njia ya Newton inahitaji kufikia kiwango cha chini zaidi?
   • Hatua kumi kabisa
   • Hatua moja
   • Nambari sawa na kushuka kwa gradient
   • Haiunganishi kamwe
3. Kwa nini njia kamili ya Newton haiwezi kutumika kwa mitandao ya neural ya parameta bilioni?
   • Upinde rangi hauwezi kukokotolewa
   • Viingilio vya pili kila wakati huwa sifuri katika mitandao ya neva
   • Hessian ni N-na-N, inagharimu kama kumbukumbu ya N-mraba na N-cubed compute ili kugeuza
   • Inafanya kazi kwenye picha tu
4. Je! njia za quasi-Newton kama BFGS hufanya nini ili kuzuia gharama ya Hessian?
   • Wanapuuza gradient kabisa
   • Wao nasibu nadhani curvature
   • Wanabadilisha shida kuwa laini
   • Wanaunda ukadiriaji wa Hessian kinyume kutoka kwa tofauti za gradient zinazofuatana
5. L-BFGS inapunguzaje kumbukumbu ikilinganishwa na BFGS?
   • Huhifadhi gradient chache tu za mwisho na vekta za hatua badala ya tumbo kamili
   • Inatumia kuelea kwa usahihi mmoja pekee
   • Inatupa upinde rangi baada ya kila hatua
   • Inatoa mafunzo kwa vigezo vichache
6. Ni mbinu ipi ya kisasa inayokadiria mzingo ili kuharakisha mafunzo makubwa ya mtandao wa neva?
   • Usimbaji wa moto-moja
   • Njia zinazotokana na Kronecker kama K-FAC na Shampoo
   • Mteremko wa upinde wa mvua wa stochastiki usio na marekebisho
   • Kuacha
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