Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Idanwo Ẹkọ Olona-ṣiṣe

Checks hard and soft task sharing, weighted objectives, negative transfer and methods for handling conflicting task gradients.

Jẹmọ itọsọna awọn ọnaẸkọ Iṣẹ-ṣiṣe pupọ
Ibeere 1 ti 8

Bawo ni awoṣe ikẹkọ iṣẹ-ọpọlọpọ ti ṣeto lakoko ikẹkọ?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Bawo ni awoṣe ikẹkọ iṣẹ-ọpọlọpọ ti ṣeto lakoko ikẹkọ?
    • Ikẹkọ awoṣe kan lori ọpọlọpọ awọn iṣẹ ṣiṣe ti o jọmọ pẹlu awọn aṣoju pinpin
    • Ikẹkọ awoṣe lọtọ fun gbogbo iṣẹ-ṣiṣe
    • Lilo data ti ko ni aami nikan fun iṣẹ-ṣiṣe kan
    • Yọ gbogbo awọn ipele ti o farapamọ kuro ni nẹtiwọki kan
  2. Ni aṣoju lile-paramita-pinpin MTL nẹtiwọki, kini o pin ati kini o jẹ lọtọ?
    • Ohun gbogbo ni lọtọ fun iṣẹ-ṣiṣe
    • Egungun ti o pin pẹlu awọn olori iṣelọpọ iṣẹ-ṣiṣe kan pato
    • Nikan iṣẹ isonu ti pin
    • Nikan data igbewọle, kii ṣe awọn fẹlẹfẹlẹ
  3. Kini idi ti ẹkọ iṣẹ-ọpọlọpọ le ṣe ilọsiwaju gbogbogbo?
    • O ṣe imukuro iwulo fun eyikeyi data
    • O ṣe onigbọwọ aṣiṣe ikẹkọ odo
    • Awọn ifihan agbara lati awọn iṣẹ ṣiṣe ti o jọmọ ṣe deede aṣoju pinpin ati dinku iwọn apọju
    • O yọ iwulo fun awọn ori iṣẹ-ṣiṣe pato
  4. Kini 'gbigbe odi' ni ẹkọ iṣẹ-ṣiṣe pupọ?
    • Nigbati fifi awọn iṣẹ-ṣiṣe kun iyara ikẹkọ
    • Nigbati awoṣe ba ni awọn paramita diẹ ju
    • Nigbati gbogbo awọn iṣẹ-ṣiṣe pin awọn aami kanna
    • Nigbati ẹkọ iṣẹ-ṣiṣe kan ṣe ipalara iṣẹ lori miiran nitori awọn ifihan agbara ikọlura
  5. Bawo ni ipadanu gbogbogbo ṣe agbekalẹ nigbagbogbo ni ikẹkọ iṣẹ-ṣiṣe pupọ?
    • Gẹgẹbi iwọn iwuwo ti awọn adanu iṣẹ-ṣiṣe kọọkan
    • Nipa gbigbe nikan pipadanu iṣẹ-ṣiṣe ti o tobi julọ
    • Nipa aibikita gbogbo rẹ ṣugbọn iṣẹ-ṣiṣe kan
    • Nipa isodipupo awọn adanu pọ
  6. Kini idi ti awọn awoṣe ede ti o tobi julọ ṣe akiyesi iṣẹ-ṣiṣe olona-ara?
    • Wọn le ṣe iṣẹ-ṣiṣe ti o wa titi kan nikan
    • Nẹtiwọọki ẹyọkan n ṣakoso itumọ, akopọ, ifaminsi, ati Q&A
    • Wọn lo awoṣe ti o yatọ fun itọka kọọkan
    • Wọn ko pin awọn paramita kọja awọn iṣẹ ṣiṣe
  7. In soft parameter sharing, how are task models usually related?
    • They use one identical model with no task-specific outputs
    • They share no information and cannot exchange features
    • They keep separate parameters but add a constraint or penalty encouraging similarity
    • They share only the training labels
  8. What does PCGrad do when two task gradients conflict?
    • It discards every task except the largest one
    • It changes the task labels
    • It replaces the shared backbone with separate datasets
    • It projects away conflicting gradient components before combining updates