quiz bu lëkkale ak tegtal · Diggante Niveau

Model Extraction ak Sàcc Cong Quiz

Xam naka la songekat yi di toppandoo xeetu black-box ci laaj yi, li API yi di wane, ak ni fournisseur yi mëna gisee wala tere génne gi.

Yooni njiit yu jëm ciModel Extraction Ak Sàcc Cong
Laaj 1 de 8

Lan mooy xalaat bu mag bi ci ginaaw attaque extraction model?

Xoolaatal laaj bu nekk8 laaj

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

  1. Lan mooy xalaat bu mag bi ci ginaaw attaque extraction model?
    • Laajte model bi ak tàggat ab wuutukat ngir toppandoo ay génnam
    • Pirate ci serwër bi ngir kopie fichier diisaay yi ci saasi
    • Xaarandi ab liggéeykat ngir mu sacc ay done yuñ tàggat
    • Sarse API bi bamu tass ba noppi sànni mémoire bi
  2. Lan moo waral delloosi vecteur probabilite yu mat yi gëna yombal génne gi moo gën delloosi benn etiketu top-1 kese?
    • Dafay encrypter tontu bi, di yeexal ki koy song
    • Dafay joxe leeral yu bari ci laaj bu nekk, kon laaj yu néew lañu soxla
    • Dafay nëbb àppu dogal bi
    • Dafay yokk njëgu serwër bi suko defee songekat yi bàyyi
  3. Ban pexem kaaraange mooy wàññi ci saasi xibaar yiy sacc ci laaj bu nekk?
    • Yokk yeneen GPU ci serwër bi
    • Yokk limu parametru model bi
    • Delloosi génnekaay yu wër wala top-1 kese
    • Siiwal done tàggat yaram
  4. What did the 2024 production-language-model extraction paper report about GPT-3.5 Turbo?
    • It reconstructs every GPT-3.5 parameter exactly for under $20
    • It obtains the model’s training dataset directly
    • It extracts a complete model checkpoint from the API
    • It recovers the embedding-projection layer; the paper estimates under $2,000 in queries for GPT-3.5 Turbo
  5. Lan moo waral xeetu weccitu buñ sàcc nekk lu jafee gis lu weesu li ñuy ñàkk ci xaalis?
    • Dafay may song white-box yi nekk ci biti, lu ci melni defar misaali noon
    • Dafay gëna baaxal njubte gi ci modelu njëkk ba
    • Dafay efaase ci saasi model bi njëkk
    • Dafay laaj ñu tàggataat orginaal bi
  6. In ActiveThief, what does the combined DFAL + K-center query strategy aim to balance?
    • Boundary-proximate informative samples and reduced redundancy among selected samples
    • Exact recovery of all secret model weights and a guaranteed substitute
    • Random sampling only and avoiding all substitute-model predictions
    • Higher API throughput and removal of the query budget
  7. What can API rate limits do against a model-extraction attempt?
    • Prove that no substitute model can be trained
    • Remove all information from each permitted prediction
    • Make large-scale querying slower or more expensive
    • Make model outputs impossible to compare
  8. What does the DynaMarks paper propose for a model-extraction defense?
    • It adds a logo to every generated answer
    • It encrypts all model weights at rest
    • It guarantees ownership of every copied production model
    • It can embed a signature in a surrogate model by dynamically changing API outputs using secret parameters