Laajte GAN bu Wasserstein

Sooy def quiz bi, tontu yi ak leeral yi dañuy feeñ.

1. Ban metric distance la WGAN di jëfandikoo ngir méngale séddale dëgg ak yiñ defar?
   • Diggante Wasserstein (kiy toxal suuf)
   • Kullback-Leibler
   • Nuru cosinus
   • Diggante pixel euclidien
2. Ci WGAN, reso bi daan xeex, ñu soppi tur wi mu tuddu lan, ak lu tax?
   • 'Attekat' ndax wote waaw mbaa deet
   • 'Critique' ndax dafay genne poñ yu wéy di am, du ay probabilite
   • 'Sampler' ndax dafay defar bruit
   • 'Filtre' ndax dafay dindi nataal yu baaxul yi
3. Lan moo tax ñu wara digal ŋàññikat WGAN mu nekk 1-Lipschitz?
   • Ngir gëna gaaw ci tàggat ci GPU yi
   • Ngir wàññi limu parametre yi
   • Ndax ñaari mbir yu Kantorovich-Rubinstein yi ñuy jëfandikoo ngir xayma diggante Wasserstein dafay laaj ay degrade yu yam
   • forse ko mu genne ay lim yu matt
4. naka la WGAN bi njëkkee tëralee li Lipschitz tëral?
   • Soo yokke dropout
   • Jëfandikoo normalisasioŋ ci lots kese
   • Ni ñuy gelée generatër bi
   • Dagg diisaayu ŋàññikat bi ci diggante bu ndaw
5. Ban jafe-jafe la WGAN gëna wàññi buñu ko méngale ak vanille GANs?
   • Mode collapse ak perte bu amul xibaar
   • Overfitting ci ensemble test bi
   • Nataal bi dafay yeex
   • Jëfandikoo memory bu bari
6. Ban yokkute la WGAN-GP indi ci WGAN bi njëkk?
   • Generatër bu gëna mag
   • Penalti gradient buy wecci dagg diisaay bi
   • Done yu bees
   • Neexal jàng buy dooleel
7. Why can Wasserstein loss provide a useful signal when real and generated distributions barely overlap?
   • It becomes a binary accuracy score
   • It no longer depends on the critic
   • The distance can vary smoothly as the distributions move, providing gradients before they overlap
   • It requires the generator to match every sample exactly
8. What does the WGAN-GP gradient penalty encourage the critic’s gradient norm to do on sampled interpolations?
   • Become exactly zero everywhere
   • Grow without bound
   • Match the generator’s parameter count
   • Stay near one