Maswali Hasi ya Sampuli na Kelele za Makadirio ya Kinyume cha Kelele

Majibu na maelezo huonekana unapojibu maswali.

1. Je, sampuli hasi na NCE hutatua tatizo gani kimsingi?
   • Gharama kubwa ya kompyuta ya softmax kamili juu ya nafasi kubwa sana ya pato
   • Kutoshea kupita kiasi kwenye hifadhidata ndogo
   • Kutoweka kwa gradient katika mitandao ya kawaida
   • Ukosefu wa data iliyo na lebo kwa masomo yanayosimamiwa
2. Je, Kadirio la Kinyume cha Kelele huwekaje upya jukumu la kujifunza?
   • Kama rejeshi kwenye malengo endelevu
   • Kama uainishaji wa binary kati ya sampuli za data za kweli na sampuli za kelele
   • Kama mkusanyiko wa pointi zisizo na lebo
   • Kama mafunzo ya kuimarisha na thawabu
3. Ni modeli gani iliyoeneza sampuli hasi za kujifunza upachikaji wa maneno?
   • BERT
   • ResNet
   • neno2vec's skip-gram mfano
   • GPT-2
4. Sampuli hasi inatofautiana vipi na NCE kamili?
   • Inaongeza masharti ya ziada ya kuhalalisha kwa usahihi
   • Inahitaji msamiati mzima katika kila hatua
   • Inatumia upotezaji wa rejista badala ya sigmoid
   • Huangusha masharti ya urekebishaji wa NCE, na kuifanya iwe haraka lakini sio kikadiriaji thabiti cha msongamano
5. Kwa nini hasi mara nyingi huchukuliwa kutoka kwa usambazaji wa unigram huinuliwa hadi nguvu 3/4?
   • Kusawazisha ushawishi wa maneno ya mara kwa mara na adimu badala ya kuchukua sampuli kulingana na marudio
   • Kupuuza maneno adimu kabisa
   • Kufanya maneno yote yawe sawa
   • Kuondoa maneno ya kuacha kutoka kwa mafunzo
6. For a large vocabulary, how can sampled negative pairs reduce work compared with evaluating every word in a full softmax?
   • It computes an exact probability for every vocabulary item
   • The update scores a positive pair and a limited set of sampled negatives instead of normalizing over the whole vocabulary
   • It eliminates the need to score the positive pair
   • It sorts all words by frequency on each step
7. In skip-gram negative sampling, what is treated as the positive training pair?
   • Two words sampled independently from the noise distribution
   • A word paired with an empty context
   • A word and a context word observed together in the corpus
   • A validation label paired with a model checkpoint
8. Why is negative sampling not generally interpreted as estimating a normalized language-model probability?
   • It always sums over the full vocabulary
   • It requires exact maximum-likelihood normalization
   • It never uses negative examples
   • Its simplified binary objective drops NCE normalization terms and targets useful embeddings instead