Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Ẹkọ Ilọsiwaju ati Idanwo Igbagbe Ajalu

Review catastrophic forgetting, stability-plasticity, EWC, replay, task families, and continual-learning limits. Check scope and assumptions.

Jẹmọ itọsọna awọn ọnaẸkọ Ilọsiwaju Ati Igbagbe Ajalu
Ibeere 1 ti 8

Kini 'igbagbe ajalu'?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini 'igbagbe ajalu'?
    • Nigbati o ba nkọ iṣẹ-ṣiṣe tuntun kan dinku iṣẹ ṣiṣe lori awọn iṣẹ ṣiṣe ti a ti kọ tẹlẹ
    • Nigbati awoṣe ba jade kuro ni iranti ati awọn ipadanu
    • Nigbati data ikẹkọ ti paarẹ lairotẹlẹ
    • Nigbati awoṣe ba kọ lati ṣe awọn asọtẹlẹ
  2. Iṣowo-pipa ni ọkan ti ẹkọ tẹsiwaju nigbagbogbo ni a pe ni:
    • Ibajẹ-iyatọ-iyatọ
    • Iduroṣinṣin-plasticity atayanyan
    • Iyara-pipe iṣowo
    • Iwakiri-iṣoro ilokulo
  3. Kini Isọdọtun Iwọn Elastic (EWC) ṣe?
    • Npa awọn neuronu ti ko ṣe pataki
    • Ṣe ilọpo meji oṣuwọn ikẹkọ fun awọn iṣẹ ṣiṣe tuntun
    • Ṣe ijiya awọn iyipada si awọn iwuwo pataki fun awọn iṣẹ ṣiṣe ti a kọ tẹlẹ
    • Tọju gbogbo apẹẹrẹ ikẹkọ ti o kọja lailai
  4. Bawo ni awọn ọna ti o da lori atunṣe ṣe koju igbagbe?
    • Nipa didi gbogbo nẹtiwọọki naa patapata
    • Nipa lilo iwọn ẹkọ ti o tobi ju
    • Nipa yiyọ iṣẹ pipadanu
    • Nipa interleaving ti o ti fipamọ tabi ti ipilẹṣẹ atijọ apẹẹrẹ sinu titun ikẹkọ
  5. How does Gradient Episodic Memory (GEM) constrain a gradient update while learning a new task?
    • It projects the gradient so its first-order approximation does not increase loss on stored prior-task examples
    • It replaces all prior-task data with new examples
    • It freezes every parameter for all future tasks
    • It increases weights that were unimportant to earlier tasks
  6. Kini idi ti iṣatunṣe itanran ti o rọrun lori iṣẹ-ṣiṣe tuntun n fa igbagbe?
    • GPU overheats ati ba iranti jẹ
    • Isọkalẹ ilọkuro n gbe awọn iwuwo pinpin si iṣẹ ṣiṣe tuntun laisi idiwọ lati tọju awọn atijọ
    • Awọn iṣẹ-ṣiṣe titun nigbagbogbo ni data diẹ sii
    • Awọn awoṣe imomose npa atijọ òṣuwọn
  7. How does iCaRL classify a test example after learning multiple classes?
    • By retrieving the single closest word in a text corpus
    • By selecting the class with the largest Fisher information
    • By comparing its feature vector with per-class means computed from stored exemplars
    • By replaying every training image at inference
  8. Why is evaluating continual learning more than measuring final accuracy on the newest task?
    • New-task accuracy is the only relevant measure
    • Old tasks are removed from the evaluation
    • The model cannot be tested on multiple tasks
    • A method must also measure retention on prior tasks and its ability to learn new ones