Sondage lineaire ak Quiz buy jàngat màndarga yiñ gelé

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

1. Ci sonde lineaire buñ miin, ban parametre lañuy yeesal ci diiru tàggat bi?
   • Classificateur lineaire bu bees biñu yokk ci kaw
   • Yaxu ndigg li yépp ak couche ligneaire bi
   • Yaxu ndigg li kese, bàyyi classifier bi
   • Layer biy dugal ak bloku bàyyi xel bu mujj bi
2. What does high linear-probe accuracy most directly show?
   • The backbone causally uses the feature for every prediction
   • The labels are decodable by a linear classifier from the frozen features under that dataset and protocol
   • The representation is robust to every distribution shift
   • The probe has identified a human-interpretable mechanism
3. Buñu ko méngale ak fine-tuning bu mat sëkk, sondage lineaire moo gëna neex ndax:
   • Dina gëna dëggu ci bépp liggéey
   • Garantie ngir yamale lu gëna baax ci bitti séddale
   • Dafa gëna yomb ci ordinatër ak ci mémoire te dafay joxe lire bu gëna sell ci kalite man-man yi
   • Benn anam biñu mëna jëfandikoo xeetu tàggat bu njëkk
4. Li gëna bari luñu koy gis suñuy sondee transformatër yi mooy:
   • Couche bu mujj bi mooy joxe sonde bi gëna dëggu
   • Sondage du dox ludul ci reso convolutionnel
   • Tanneef layer du musa am njeexital ci resultaa yi
   • Diggante couche yi dañuy faral di gëna mëna toxal couche yu mujj yi
5. Ginaaw benchmarking, sonde lineaire yi ñu ngi leen di gëna jëfandikoo ngir tekki:
   • Saytu ndax ab konsept bu melni dëgg defa enkode ci anam yu nëbbu
   • Modèlu kompresse ngir jëfandikoo mobile
   • Wecci optimisatër bi ci diiru tàggat yaram
   • Defar ci saasi done tàggat yu bees
6. Risk bi gëna mag ci wàllu njuréef sooy tekki ab sondaas bu jaar yoon mooy:
   • Yaxu ndigg li dafay wéy di tàggat ci nëbbëtu
   • Sondaas bi mën na nekk muy jëfandikoo ab korrelaasioŋ bu dul dëgg moo gën konsept bu dëggu bi
   • Sonde lineaire mënuñu ëpp
   • Mënu ñu jëfandikoo entropie cross
7. To compare two backbone layers with linear probes, what should remain consistent?
   • The feature vectors from each layer
   • The layer being evaluated
   • The task labels, data split and probe-training protocol
   • The activation values themselves
8. Why might feature normalization and learning-rate sweeps be used in a probe evaluation?
   • To update the frozen backbone
   • To add hidden nonlinear layers to the probe
   • To guarantee the concept is causally used
   • To reduce optimization and scale effects that could distort the representation comparison