Bottleneck Architectures adanwo

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

1. Ninu bulọọki igo ResNet kan, kini ipa ti convolution 1x1 akọkọ?
   • Lati din awọn nọmba ti awọn ikanni ṣaaju ki o to 3x3 convolution
   • Lati mu ipinnu aaye pọ si
   • Lati waye ik classification
   • Lati fi ariwo kun fun deede
2. Kini idi ti igo kan jẹ ki awọn nẹtiwọọki ti o jinlẹ din owo lati ṣe iṣiro?
   • O yọ gbogbo awọn iṣẹ ṣiṣe kuro
   • O ṣe awọn iṣẹ ṣiṣe gbowolori lori awọn ikanni diẹ
   • O nlo awọn kernel ti o tobi julọ nibi gbogbo
   • O skips backpropagation
3. Ninu ohun autoencoder, Layerneck Layer pinnu nipataki:
   • Aṣayan optimizer
   • Ọna kika faili aworan
   • Elo ni igbewọle ti wa ni fisinuirindigbindigbin
   • Iwọn ipele naa
4. MobileNetV2 nlo iyatọ wo ti imọran igo?
   • Pure pooling bottleneck
   • Loorekoore igo
   • Ko si bottleneck ni gbogbo
   • Ọrun igo to ku ti o yipada (faagun lẹhinna adehun)
5. Bawo ni LoRA ṣe lo awọn ilana igo si awọn awoṣe ede nla?
   • Nipa fifi sii igo-igo kekere ti o le kọni sinu awọn imudojuiwọn iwuwo
   • Nipa piparẹ awọn ipele akiyesi
   • Nipa ilọpo meji gbogbo awọn iwuwo
   • Nipa pipọ si 1 bit
6. A aṣoju ResNet bottleneck Àkọsílẹ wọnyi eyi ti ikanni Àpẹẹrẹ?
   • Gbooro, fife, fife
   • Din, ilana, mu pada (fun apẹẹrẹ, 256 si 64 si 256)
   • Din, gbooro, dín
   • ID fun Layer
7. A ResNet block reduces 256 channels to 64 before its 3×3 convolution. Ignoring the 1×1 layers, what is the 3×3 layer’s approximate multiply-add cost relative to 256 channels?
   • About one-quarter
   • About one-half
   • About one-sixteenth
   • About four times as much
8. An autoencoder uses a narrow latent code. What constraint does this impose on reconstruction?
   • The encoder must preserve every input pixel exactly
   • The network can no longer use learned weights
   • The decoder must expand the latent code before encoding
   • The decoder must reconstruct the input from a compressed representation