Kediska Ururinta Gradient

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Muxuu ururinta gradient ugu horrayn kuu ogolaanayaa inaad samayso?
   • Tareen la'aanteed wax GPU ah
   • Ka bood baaska dambe
   • Ku ekaysii cabbir weyn oo waxtar leh oo ku saabsan xusuusta xaddidan
   • Tirtir guud ahaan wax-qabadka
2. Iyada oo cabbirka-yar-yar ee 4 iyo 8 tillaabooyinka ururinta, cabbirka dufcadda waxtarka leh waa:
   • 32
   • 4
   • 8
   • 12
3. With an accumulation factor N, when should optimizer.step() and zero_grad() occur?
   • Ka dib koob kasta oo yar yar
   • Marna
   • Ka hor inta aan hore u dhaafin
   • Call optimizer.step() after N micro-batches, then clear gradients before the next accumulation window
4. Si heerka waxbarashada waxtarka leh loo ilaaliyo mid sax ah, khasaaraha dufcaddii-yaro-yar inta badan waa:
   • Lagu dhuftey N
   • Waxaa loo qaybiyaa tirada tillaabooyinka ururinta N
   • U dhig eber
   • Lagu bedelay si sax ah
5. Waa kuwee lakabka caadiga ah ee si nadiif ah ugu lammaaniya ururinta gradient?
   • Lakab ama koox caadi ah, maadaama aysan ku xirneyn tirakoobka dufcada
   • Dufcaddii caadiga ahayd, sababtoo ah waxay jeceshahay dufcooyin yaryar
   • Ma jiro wax caadi ah oo la shaqeeya
   • Kaliya ka tagista
6. Qiimaha ugu muhiimsan ee isticmaalka raasamaal gradient waa:
   • Xusuusta GPU badan oo la isticmaalo marka ugu sarreysa
   • Luminta dhammaan jaranjarooyinka
   • Wakhti dheer oo saacadda gidaarka ah cusboonaysiin kasta oo waxtar leh
   • Awood la'aanta isticmaalka hagaajinta
7. A training loop clears gradients after every micro-batch but steps the optimizer every fourth batch. What is lost?
   • The model weights are deleted
   • The earlier micro-batch contributions to the accumulated update
   • The effective batch becomes four times larger
   • The loss automatically becomes zero
8. Why may four micro-batches with BatchNorm behave differently from one physical batch of the same total size?
   • The optimizer cannot accept four backward passes
   • Gradient addition is forbidden in BatchNorm
   • The number of training examples changes
   • BatchNorm uses statistics from each micro-batch rather than the combined batch