Maswali ya Wastani wa Uzito wa Stochastic

Majibu na maelezo huonekana unapojibu maswali.

1. Je, wastani wa Uzito wa Stochastic ni wastani gani?
   • Uzito wa modeli uliyotolewa katika sehemu nyingi marehemu katika mafunzo
   • Utabiri wa mifano kadhaa iliyofunzwa tofauti
   • Gradients zilikokotoa kwa bachi zote ndogo
   • Viwango vya kujifunza vilivyotumika katika kila enzi
2. Kwa nini SWA ina mwelekeo wa kuboresha jumla?
   • Inakariri mafunzo yaliyowekwa kwa usahihi zaidi
   • Hupata minima pana, bapa ambayo ni thabiti zaidi kwenye data isiyoonekana
   • Inaongeza idadi ya vigezo katika mfano
   • Huondoa hitaji la kuweka uthibitishaji
3. Je, ni hatua gani ya ziada ambayo SWA inahitaji kwa mitandao inayotumia urekebishaji wa bechi?
   • Kuunda upya mtandao kutoka mwanzo
   • Kuongeza ukubwa wa kundi mara mbili
   • Pasi ya ziada ya mbele ili kurudisha takwimu za uendeshaji wa BN
   • Kuondoa tabaka zote za kuacha
4. Ni tabia gani ya kiwango cha ujifunzaji hutumika kwa kawaida wakati wa awamu ya wastani ya SWA?
   • Kiwango cha kujifunza kimeshuka hadi sifuri haswa
   • Kiwango cha kujifunza cha moja kwa utulivu
   • Hakuna kiwango cha kujifunza hata kidogo kwani mafunzo yamegandishwa
   • Kiwango cha kawaida au cha mzunguko, cha juu kiasi cha kujifunza
5. Gharama ya uelekezaji wa SWA inalinganishwaje na mkusanyiko wa jadi wa mifano ya N?
   • Inagharimu takriban mfano mmoja, kwani matokeo ni seti moja ya uzani
   • Ni mara N ghali zaidi
   • Ni sawa na kuendesha mifano ya N
   • Haiwezi kutumika kwa hitimisho
6. Why does the batch-normalization adjustment belong after SWA weight averaging?
   • The averaging step deletes every convolution
   • Its running statistics must be recomputed for the averaged weights
   • It converts batch normalization into layer normalization
   • It restores the original optimizer learning rate
7. An SWA running mean combines n previous samples with a new weight vector w_i. Which update gives equal weight to all n+1 samples?
   • w_SWA + n · w_i
   • (w_SWA + w_i) divided by n
   • (n · w_SWA + w_i) divided by (n + 1)
   • n · (w_SWA − w_i)
8. Why is SWA commonly started during the later part of training?
   • It requires the model to remain at random initialization
   • It can only average models with different architectures
   • It replaces the training data with validation examples
   • It averages nearby useful checkpoints from a trajectory after the model has made substantial progress