Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Sitokasitik àdánù aropin adanwo

Checks SWA weight averaging, learning-rate behavior, batch-normalization recalibration and how a single averaged model differs from an ensemble.

Jẹmọ itọsọna awọn ọnaSitokasitik iwuwo aropin
Ibeere 1 ti 8

Kini Iwọn Iwọn Sitokasitik gangan ni aropin?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini Iwọn Iwọn Sitokasitik gangan ni aropin?
    • Awọn iwuwo ti awoṣe ti a ṣe ayẹwo ni awọn aaye pupọ pẹ ni ikẹkọ
    • Awọn asọtẹlẹ ti ọpọlọpọ awọn awoṣe ikẹkọ lọtọ
    • Awọn gradients ṣe iṣiro kọja gbogbo awọn ipele kekere
    • Awọn oṣuwọn ikẹkọ ti a lo lakoko akoko kọọkan
  2. Kini idi ti SWA maa n mu ilọsiwaju pọ si?
    • O ṣe akori eto ikẹkọ diẹ sii ni pipe
    • O rii gbooro, minima fifẹ ti o lagbara diẹ sii lori data ti a ko rii
    • O mu ki awọn nọmba ti sile ni awọn awoṣe
    • O ṣe imukuro iwulo fun eto afọwọsi kan
  3. Igbese afikun wo ni SWA nilo fun awọn nẹtiwọọki ti o lo isọdọtun ipele?
    • Retraining awọn nẹtiwọki lati ibere
    • Ilọpo meji iwọn ipele
    • Afikun siwaju kọja lati tun ṣe iṣiro awọn iṣiro ṣiṣiṣẹ BN
    • Yiyọ gbogbo dropout Layer
  4. Iwa-oṣuwọn ẹkọ wo ni a lo nigbagbogbo lakoko ipele aropin SWA?
    • Oṣuwọn ẹkọ ti bajẹ si odo gangan
    • Oṣuwọn ẹkọ ti ọkan fun iduroṣinṣin
    • Ko si oṣuwọn ikẹkọ rara nitori ikẹkọ ti di tutu
    • Iduroṣinṣin tabi iyipo, oṣuwọn ẹkọ ti o ga julọ
  5. Bawo ni iye owo idawọle SWA ṣe afiwe si akojọpọ ibile ti awọn awoṣe N?
    • O-owo ni aijọju awoṣe ẹyọkan, nitori abajade jẹ ṣeto awọn iwuwo kan
    • O ti wa ni N igba diẹ gbowolori
    • O jẹ aami kanna si ṣiṣe awọn awoṣe N
    • Ko le ṣee lo fun itọkasi
  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