Idanwo ti o sopọ mọ itọsọna · Rọrun Ipele

Idanwo Augmentation Data

Review data augmentation methods for images and text, including label-preserving transformations, mixing and learned policies.

Jẹmọ itọsọna awọn ọnaAugmentation data
Ibeere 1 ti 8

Kini idi akọkọ ti imudara data?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini idi akọkọ ti imudara data?
    • Lati faagun ati ṣe iyatọ data ikẹkọ ki awọn awoṣe ṣe gbogbogbo dara julọ
    • Lati titẹ soke inference on a oṣiṣẹ awoṣe
    • Lati compress iwọn ti dataset lori disk
    • Lati ṣe aami awọn apẹẹrẹ ti ko ni aami laifọwọyi
  2. Ewo ninu iwọnyi jẹ ilana imudara aworan ti o wọpọ?
    • Isosile ite
    • Iyipada petele ID
    • Àmi
    • Jade kuro
  3. Kini itumọ-pada ṣe ni imudara ọrọ?
    • Yipada ilana ti awọn ọrọ ni gbolohun ọrọ kan
    • Yọ gbogbo awọn ọrọ idaduro kuro ninu ọrọ naa
    • Tumọ ọrọ si ede miiran ati pada lati ṣẹda awọn atumọ
    • Iyipada ọrọ sinu awọn ifibọ nomba
  4. Kini idi ti awọn afikun gbọdọ jẹ 'itọju-aami-aami'?
    • Lati tọju iwọn faili dataset nigbagbogbo
    • Lati jẹ ki ikẹkọ ṣiṣe ni iyara
    • Nitori awọn awoṣe ko le ka awọn akole bibẹẹkọ
    • Nitori iyipada aami yoo kọ awọn ẹgbẹ aṣiṣe awoṣe
  5. Kini ilana Mixup ṣe?
    • Papọ awọn apẹẹrẹ meji ati awọn aami wọn sinu apẹẹrẹ ikẹkọ kan
    • Awọn iboju iparada agbegbe onigun ti aworan kan
    • Ṣewadii eto imulo imudara ti o dara julọ laifọwọyi
    • Ṣe afikun ariwo laileto si awọn iwuwo awoṣe
  6. Nigbawo ni imudara data paapaa niyelori?
    • Nigbati o ti ni data aami ailopin tẹlẹ
    • Nigbati data ike ba ṣọwọn ati gbigba diẹ sii jẹ lile
    • Nikan nigbati awoṣe ba ni ikẹkọ ni kikun
    • Nikan fun data iwe kaunti tabular
  7. What does CutMix do to an image and its training label?
    • It adds a random word to a sentence without changing the label
    • It swaps only class names in the dataset
    • It replaces a patch with a patch from another image and mixes labels by the replaced area
    • It blurs every input and keeps a single label unchanged
  8. How does AutoAugment select augmentation policies in its original paper?
    • It chooses transformations from a fixed human-written list without evaluation
    • It uses test labels to relabel training images
    • It selects operations from inference-time confidence only
    • It searches for policies using validation performance as a reward signal