Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Awọn Nẹtiwọọki Siamese ati Idanwo Ipadanu Triplet

Checks shared encoders, embedding distances, triplet margins and how hard-negative examples affect the learning signal. carefully.

Jẹmọ itọsọna awọn ọnaAwọn nẹtiwọki Siamese Ati Pipadanu Meta
Ibeere 1 ti 8

Kini ẹya ara ẹrọ asọye ti nẹtiwọọki Siamese kan?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini ẹya ara ẹrọ asọye ti nẹtiwọọki Siamese kan?
    • Meji tabi diẹ ẹ sii aami ẹka ti o pin kanna òṣuwọn
    • Ẹka ẹyọkan ti ko si awọn fẹlẹfẹlẹ ti o farapamọ
    • Awọn nẹtiwọọki lọtọ pẹlu awọn iwuwo ominira patapata
    • Loop loorekoore lori awọn igbesẹ akoko
  2. Ni ipadanu meteta, kini awọn igbewọle mẹta naa?
    • Iṣagbewọle, farasin, iṣẹjade
    • Oran, rere, ati odi
    • Reluwe, afọwọsi, igbeyewo
    • Encoder, decoder, classifier
  3. Kini ipinnu ipadanu meteta ṣe imuṣẹ?
    • Oran ati odi yẹ ki o jẹ aami
    • Gbogbo awọn ifibọ yẹ ki o ṣubu si aaye kan
    • Ijinna idarọ-rere yẹ ki o kere ju ijinna oran-odi lọ nipasẹ ala kan
    • Rere ati odi yẹ ki o ni aaye dogba si oran
  4. Kini idi ti 'iwakusa odi lile' ṣe pataki fun ikẹkọ ipadanu meteta?
    • O yọ gbogbo rere kuro lati ikẹkọ
    • O mu ki awọn nẹtiwọki aijinile
    • O rọpo metiriki ijinna pẹlu aami kan
    • Awọn odi ti o rọrun ti ni itẹlọrun ala tẹlẹ ati gbejade gradient iwulo odo
  5. Agbara wo ni o jẹ ki awọn nẹtiwọọki Siamese ṣe pataki fun 'ẹkọ-ibọn-ọkan’?
    • Wọn le ṣe afiwe igbewọle tuntun si itọkasi ẹyọkan ati idajọ ibajọra, paapaa fun awọn kilasi ti a ko rii
    • Wọn nilo ẹgbẹẹgbẹrun awọn apẹẹrẹ aami fun kilasi kan
    • Wọn pin nikan si awọn ẹka ti o wa titi, ti a ti sọ tẹlẹ
    • Wọn ko le ṣe gbogbogbo kọja awọn idamọ ikẹkọ
  6. Bawo ni nẹtiwọọki Siamese ṣe pinnu igbagbogbo ti awọn igbewọle meji ba baamu?
    • Nipa apapọ awọn iye piksẹli aise wọn
    • Nipa ifiwera aaye laarin awọn ifisinu ifisinu wọn lodi si iloro kan
    • Nipa kika awọn nọmba ti fẹlẹfẹlẹ lo
    • Nipa fifun kọọkan aami kilasi laileto
  7. For triplet loss max(0, d(a,p) − d(a,n) + m), let d(a,p)=0.2, d(a,n)=0.8 and m=0.3. What is the loss?
    • 0.3
    • 0.5
    • 0
    • 1.1
  8. What happens when a negative example is already farther from the anchor than the required margin?
    • The anchor and positive embeddings are discarded
    • The loss forces the negative closer to the anchor
    • The network switches to classification mode
    • The triplet loss becomes zero and that example gives no useful gradient