Iṣapẹẹrẹ Iṣeto ati Iyatọ Iyatọ Idanwo

Awọn idahun ati awọn alaye han bi o ṣe n mu ibeere naa.

1. Iyatọ ifihan dide nitori awọn awoṣe ti a ti kọ pẹlu ipa olukọ ni ipo nikan lori:
   • Ilẹ-otitọ àmi, sugbon gbọdọ lo ara wọn ni inference
   • Awọn ami asọtẹlẹ tiwọn
   • Awọn ami ariwo laileto
   • Future àmi
2. Kini iṣapẹẹrẹ iṣeto ṣe lakoko ikẹkọ?
   • Nigbagbogbo ifunni aami otitọ
   • Laileto yan laarin ami otitọ ati asọtẹlẹ ti ara awoṣe
   • Mu iṣẹ pipadanu ṣiṣẹ lorekore
   • Ṣe ilọpo meji oṣuwọn ikẹkọ lori akoko
3. Bawo ni iṣeeṣe ti lilo aami-otitọ ilẹ ṣe yipada lori ikẹkọ ni iṣapẹẹrẹ iṣeto?
   • O duro nigbagbogbo
   • Ti o ba bẹrẹ kekere ati ki o pọ
   • O bẹrẹ ga ati ibajẹ si odo
   • O oscillates laileto
4. Tani ṣe agbekalẹ iṣapẹẹrẹ iṣeto, ati ni aijọju nigbawo?
   • Vaswani ati al., Ọdun 2017
   • Hinton ati al., Ọdun 2006
   • LeCun ati al., 1998
   • Bengio et al., Ọdun 2015
5. Idinku imọ-ẹrọ ti a mọ ti ifunni awọn ami apere pada lakoko ikẹkọ ni pe:
   • Igbesẹ iṣapẹẹrẹ ọtọtọ ko ṣe iyatọ, idiju gradients
   • O jẹ ki ikẹkọ ni kikun ipinnu
   • Ko nilo data afọwọsi
   • O ṣe imukuro iwulo fun awọn ifibọ
6. Apẹrẹ iṣeto wo ni a tọka si fun ibajẹ iṣeeṣe ti ipa olukọ?
   • Ibakan
   • Iyatọ-sigmoid ibajẹ
   • Igbesẹ-soke laini ilosoke
   • Aileto rin
7. At a decoding step where scheduled sampling chooses the model-output branch, what is fed as the next context token?
   • The next ground-truth token every time
   • A token from a separate validation sequence
   • A token sampled from the model’s predicted distribution
   • The average of all vocabulary embeddings
8. Why does scheduled sampling usually begin with a high probability of using ground truth?
   • It guarantees unbiased gradients
   • It removes the need for a training schedule
   • It makes inference use ground-truth tokens
   • Early predictions are poor, so a correct prefix can stabilize initial learning