Kediska Jarida Gradient

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

1. Muxuu gooyn-si-caadiyan-gooynta ugu horrayn u ilaalinayaa iyadoo la xaddidayo cusboonaysiinta?
   • Jihada cusboonaysiinta gradient guud
   • Baaxadda saxda ah ee gradient kasta
   • Tirada qiyaasaha moodeelka
   • Jadwalka heerka waxbarashada
2. Sida caadiga ah clip-by-caadiga ah ee leh marinka c, maxaa dhacaya marka heerka gradient g_norm uu dhaafo c?
   • Gradients ayaa lagu dejiyay eber
   • Qalab kasta waxaa lagu dhuftey c/g_norm
   • Tababarku wuu joogsadaa
   • Heerka waxbarashadu waa labanlaab
3. Dhibaato noocee ah ayaa si tartiib tartiib ah loo gooyay si gaar ah loogu talagalay in lala dagaallamo?
   • Qaybaha baaba'aya
   • Soo dejinta xogta oo gaabis ah
   • Jaangooyooyin qarxa
   • Ku-habboonaanta habka ansaxinta
4. Sidee buu-qiimuhu uga duwan yahay kilib-ilaa-caadi?
   • Had iyo jeer waxay ilaalisaa jihada cusboonaysiinta
   • Waxa kaliya oo ay ka shaqeysaa lakabyo iswaafaqsan
   • Waxay ka saartaa gebi-ahaantii
   • Waxa ay si madax banaan ugu xidhaa qayb kasta oo gradient ah, taas oo bedeli karta jihada
5. A minibatch produces an unusually large global gradient norm. What does clip-by-norm do before the optimizer update?
   • Scales the full gradient vector down to the configured threshold when the norm exceeds it
   • Changes the model’s labels
   • Drops every gradient component independently
   • Increases the learning rate until the loss recovers
6. Tababarka gaarka ah ee kala duwan (DP-SGD), doorkee buu ka ciyaaraa jarista gradient-ka tusaale ahaan?
   • Waxay siraysaa xogta
   • Waxay xadaynaysaa tusaale kasta saamayntiisa sidaa darteed qaylada la cabbiray waxay dammaanad qaadi kartaa sirta
   • Waxay kordhisaa xajmiga dufcada
   • Waxay meesha ka saartaa baahida shaqo lumis
7. The global gradient norm is twice a clipping threshold c. By what factor is the gradient scaled?
   • Laba
   • One quarter
   • One half (c divided by 2c)
   • It is left unchanged
8. Why does DP-SGD clip each example’s gradient before adding calibrated noise?
   • To make all examples have the same label
   • To remove the need for a privacy budget
   • To increase the norm of rare-class gradients
   • To bound an individual example’s influence on the released update