Idanwo ti o sopọ mọ itọsọna · Alabọde Ipele

Fun pọ-ati-Excitation Networks adanwo

Checks the squeeze and excitation stages, channel reweighting, bottleneck dimension and the evidence behind the SE architecture.

Jẹmọ itọsọna awọn ọnaFun pọ Ati simi Awọn nẹtiwọki
Ibeere 1 ti 8

Kini igbesẹ 'fun pọ' ni bulọọki SE kan ṣe?

Awotẹlẹ gbogbo ibeere8 ibeere

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

  1. Kini igbesẹ 'fun pọ' ni bulọọki SE kan ṣe?
    • Nlo apapọ apapọ apapọ agbaye lati ṣe akopọ ikanni kọọkan sinu iye kan
    • Ge aworan naa si iwọn ti o kere
    • Yọ awọn ikanni laiṣe kuro patapata
    • Waye softmax lori awọn ipo aye
  2. Kini iṣejade igbesẹ 'imurasilẹ'?
    • A titun ṣeto ti convolution Ajọ
    • Iwọn ikanni kan laarin 0 ati 1
    • A bounding apoti
    • Iboju ipin kan
  3. Bawo ni a ṣe lo awọn iwuwo iwuri si awọn maapu ẹya atilẹba?
    • Fi kun eroja-ọlọgbọn
    • Ti ṣajọpọ bi awọn ikanni titun
    • Opo ikanni-ọlọgbọn
    • Lo lati boju jade awọn agbegbe aye
  4. Kilode ti igbadun naa nlo igo kan pẹlu ipin idinku (nigbagbogbo 16)?
    • Lati mu awọn nọmba ti o wu kilasi
    • Lati yọ awọn nilo fun convolutions
    • Lati jẹ ki nẹtiwọki wa ni aijinile
    • Lati tọju awọn paramita ti a ṣafikun ati ṣe iṣiro kekere
  5. Idije wo ni SENet ṣẹgun ni ọdun 2017?
    • Ipenija ipinya ImageNet ILSVRC
    • Ipenija wiwa COCO
    • Idije Kaggle Titanic
    • Aami ipilẹ GLUE
  6. Eyi ti ayaworan nigbamii sabe SE ohun amorindun fun ṣiṣe?
    • BERT ati GPT
    • EfficientNet ati MobileNetV3
    • Word2Vec ati GloVe
    • WaveNet ati Tacotron
  7. In the excitation step, which activations connect the two fully connected layers in the guide’s SE block?
    • Softmax followed by ReLU
    • Tanh followed by max pooling
    • ReLU followed by sigmoid after the second projection
    • GELU followed by layer normalization
  8. An SE block has C = 256 channels and reduction ratio r = 16. What is the bottleneck width?
    • 4 units
    • 32 units
    • 256 units
    • 16 units