Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Adanwo Igbagbe Ajalu

Study why sequentially learning new tasks can degrade previous skills and how continual-learning methods address the tradeoff.

Jẹmọ itọsọna awọn ọnaIgbagbe Ajalu
Ibeere 1 ti 8

Ipo wo ni o ṣapejuwe igbagbe ajalu ni nẹtiwọọki nkankikan?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Ipo wo ni o ṣapejuwe igbagbe ajalu ni nẹtiwọọki nkankikan?
    • Nigbati kikọ iṣẹ-ṣiṣe tuntun kan fa awoṣe lati padanu awọn agbara ikẹkọ tẹlẹ
    • Nigbati awoṣe ba jade ni iranti lakoko itọkasi
    • Nigbati data ikẹkọ ti paarẹ lairotẹlẹ
    • Nigbati awoṣe ba kọ lati dahun ibeere kan
  2. Kini idi ti igbagbe ajalu n ṣẹlẹ ni awọn nẹtiwọọki nkankikan?
    • Nitori iṣẹ-ṣiṣe kọọkan n gba nẹtiwọọki lọtọ tirẹ laifọwọyi
    • Nitori awọn imudojuiwọn gradient tuntun tun kọ awọn iwuwo pinpin ti o ṣafikun awọn ọgbọn atijọ
    • Nitoripe awọn awoṣe ko le fipamọ diẹ ẹ sii ju otitọ kan lọ
    • Nitori awọn fẹlẹfẹlẹ akiyesi pa data lori idi
  3. Kini Isọdọtun Iwọn Elastic (EWC) ṣe lati dinku igbagbe?
    • Paarẹ data iṣẹ-ṣiṣe tuntun naa
    • Ilọpo meji oṣuwọn ẹkọ
    • Ṣe ijiya awọn iyipada si awọn iwuwo pataki fun awọn iṣẹ ṣiṣe ti a kọ tẹlẹ
    • Yọ akiyesi lati awoṣe
  4. Bawo ni iriri tun ṣe (atunṣe) ija gbagbe?
    • Nipa didi gbogbo awoṣe patapata
    • Nipa ikẹkọ nikan lori data tuntun
    • Nipa idinku awọn nọmba ti fẹlẹfẹlẹ
    • Nipa interleaving ti o ti fipamọ atijọ apeere nigba titun ikẹkọ
  5. What does LoRA do to base-model weights when adding a task-specific adapter?
    • It freezes base weights and trains a low-rank adapter, which may reduce direct interference but does not guarantee retention
    • It rewrites every base weight for each new task
    • It stores no task-specific parameters
    • It guarantees zero forgetting on every task sequence
  6. Báwo ni ìmúpadàbọ̀-ọ̀nà ìmúgbòòrò ṣe ń gbàgbé àjálù?
    • Nipa piparẹ imo atijọ lati ṣe yara
    • Nipa titoju imo titun ni ohun ita database dipo ti awọn òṣuwọn
    • Nipa ikẹkọ continuously lori ID ariwo
    • Nipa piparẹ iranti awoṣe patapata
  7. How does Gradient Episodic Memory (GEM) constrain learning on a new task?
    • It removes old task examples from memory
    • It changes the model to a nearest-class-mean classifier
    • It adjusts the gradient so the first-order approximation does not increase loss on stored prior-task examples
    • It adds a KL penalty to every parameter regardless of task
  8. How do RAG-Token and RAG-Sequence differ in the original RAG paper?
    • RAG-Token removes retrieval while RAG-Sequence removes generation
    • RAG-Sequence uses images while RAG-Token uses audio
    • RAG-Token trains without model parameters
    • RAG-Token marginalizes retrieved documents per output token, while RAG-Sequence uses one document for a whole sequence