Àdánwò Initialization àdánù

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

1. Kini idi ti fifisilẹ gbogbo awọn iwuwo si odo jẹ aṣiṣe apaniyan?
   • Gbogbo awọn neuron gba awọn gradients kanna ati pe ko ṣe iyatọ rara (symmetry ko baje)
   • O jẹ ki ikẹkọ yarayara
   • O fa pipadanu lati di odi
   • O nlo iranti pupọ
2. O si (Kaiming) initialization ti wa ni pataki apẹrẹ fun eyi ti ibere ise iṣẹ?
   • Sigmoid
   • ReLU
   • Tanh
   • Softmax
3. Kini ibi-afẹde akọkọ ti awọn eto ipilẹṣẹ iwuwo ipilẹ?
   • Lati jẹ ki gbogbo awọn iwọn dogba
   • Lati ṣeto gbogbo iwuwo si ọkan
   • Lati tọju iyatọ ti awọn iṣiṣẹ ati awọn gradients ni aijọju ibakan kọja awọn fẹlẹfẹlẹ
   • Lati se imukuro awọn nilo fun data
4. Ibẹrẹ Xavier (Glorot) dara julọ fun awọn nẹtiwọọki nipa lilo awọn iṣẹ ṣiṣe wo?
   • ReLU ati awọn iyatọ rẹ
   • Awọn iṣiṣẹ laini nikan
   • Ko si ibere ise ni gbogbo
   • Tanh ati sigmoid
5. Kini idi ti O bẹrẹ ipilẹṣẹ lo iyatọ ti 2 / fan_in ju 1 / fan_in?
   • Lati sanpada fun eto ReLU ni aijọju idaji awọn igbewọle rẹ si odo
   • Lati ṣe awọn iwuwo ti o tobi ju laisi idi
   • Lati din awọn nọmba ti sile
   • Lati baramu iwọn ipele
6. Bawo ni awọn aiṣedeede ṣe ipilẹṣẹ nigbagbogbo nigbati awọn iwuwo jẹ ipilẹṣẹ laileto?
   • Si awọn iye rere ti o tobi pupọ
   • Si odo, niwon symmetry ti bajẹ tẹlẹ nipasẹ awọn iwọn laileto
   • Si iye kanna bi oṣuwọn ẹkọ
   • Si awọn iye laileto ti o tobi ju awọn iwuwo lọ
7. Under Xavier initialization, which quantities determine the weight variance in the guide’s formula?
   • The number of training epochs and batch size
   • The output class count alone
   • Fan-in and fan-out
   • The activation’s test accuracy
8. For a ReLU layer with fan-in 64, what variance does the He formula 2/fan_in specify?
   • 1/64
   • 1/16
   • 2/64 squared
   • 1/32