Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Idanwo Distillation Imọ

Apply teacher distributions, softened targets, and student evaluation in distillation, including why imitation does not guarantee teacher-level accuracy.

Jẹmọ itọsọna awọn ọnaDistillation imo
Ibeere 1 ti 8

In knowledge distillation, what is the student model trained to do?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. In knowledge distillation, what is the student model trained to do?
    • Lati ṣe ikẹkọ awoṣe ọmọ ile-iwe kekere kan lati farawe apẹẹrẹ olukọ nla kan
    • Lati ṣe awoṣe ti o tobi ati deede
    • Lati aami data laifọwọyi
    • Lati encrypt a awoṣe ká òṣuwọn
  2. Kí ni ‘ìmọ̀ òkùnkùn’ olùkọ́ túmọ̀ sí?
    • Farasin iwa irira ni awoṣe
    • Awọn ìpàrokò ìka ti awọn awoṣe faili
    • Alaye ti o wa ninu awọn iṣeeṣe ibatan ti olukọ fi si awọn kilasi ti ko tọ
    • Data olukọ ko ti kọ ẹkọ lori
  3. Ipa wo ni iwọn otutu (T) ṣe ni distillation?
    • O nṣakoso iyara ikẹkọ awoṣe nikan
    • O rọ softmax ki awọn ibajọra laarin kilasi kekere di kikọ
    • O ṣeto nọmba awọn ipele ti ọmọ ile-iwe
    • O ṣe iwọn iṣelọpọ ooru ti GPU
  4. Awọn Ayebaye distillation pipadanu daapọ eyi ti meji irinše?
    • Meji lọtọ ere awọn awoṣe
    • Itumọ asise onigun mẹrin ati pipadanu mitari kan
    • Ijiya KL kan lodi si eto imulo itọkasi ati iṣẹ iye kan
    • Oro iyatọ-KL kan ti o baamu awọn abajade rirọ ti olukọ pẹlu agbelebu-entropy lori awọn aami otitọ
  5. Ewo ni apẹẹrẹ ti awoṣe ede distilled?
    • ResNet-50
    • Ọrọ2Vec
    • DistilBERT
    • AlexNet
  6. Bawo ni distillation ti o da lori ẹya ṣe yatọ si ọna ti o da lori iṣelọpọ Ayebaye?
    • O ṣe ibaamu awọn aṣoju agbedemeji ti o farapamọ-Layer, kii ṣe awọn abajade ipari nikan
    • O kọ olukọ silẹ patapata
    • O nlo awọn akole lile nikan
    • O kọ olukọ dipo ọmọ ile-iwe
  7. A teacher assigns similar small probabilities to wolf and fox when predicting dog. What can soft-target distillation transfer beyond the top label?
    • A guarantee that either alternative is the correct label
    • The teacher’s complete original training corpus
    • The teacher’s exact number of hidden layers
    • Relative relationships among the teacher’s alternative class scores
  8. A distilled student is smaller than its teacher. Which evaluation is needed before deployment?
    • Assume its accuracy must equal the teacher’s
    • Measure its task accuracy and runtime on the intended system
    • Assume fewer parameters guarantee lower latency on every device
    • Only compare the models’ names