Linear Probing uye Frozen Feature Evaluation Quiz

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

1. Mune yakajairwa mutsara probe, ndeapi maparamita anogadziridzwa panguva yekudzidziswa?
   • Chete chichangobva kuwedzerwa linear classifier pamusoro
   • Iyo yese musana pamwe neiyo linear layer
   • Chete musana, uchisiya classifier yakagadziriswa
   • Iyo embedding layer uye yekupedzisira yekutarisisa block
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. Kuenzaniswa nekunyatso-tuning, mutsara kuongorora kunonyanya kukwezva nekuti ndiko:
   • Zvakanyanya kurongeka pabasa rese
   • Yakavimbiswa kuita generalize zvirinani kunze kwekugoverwa
   • Yakachipa mukombuta uye ndangariro uye inopa yekuchenesa kuverenga pamhando yemhando
   • Iyo chete nzira yekushandisa pretrained modhi
4. Chinhu chakajairika kuwana empirical kana uchiongorora ma transformer ndechekuti:
   • Iyo yekupedzisira layer inogara ichipa yakanakisa probe kurongeka
   • Kuongorora kunoshanda chete pane convolutional network
   • Layer sarudzo haimbokanganisa maprobe mhedzisiro
   • Matunhu epakati anowanzopfuura chikamu chekupedzisira chekutamisa
5. Kupfuura mabhenji, mutsara probes ari kuwedzera kushandiswa mukududzira kune:
   • Ziva kana pfungwa yakaita seyechokwadi yakavharwa munzvimbo dzakavanda
   • Compress modhi yekutumira nhare
   • Tsiva iyo optimizer panguva yekudzidzira
   • Gadzira data idzva rekudzidziswa otomatiki
6. Njodzi yakakosha yemaitiro kana uchidudzira yakakwira probe kurongeka ndeyekuti:
   • Musana unochengeta pakavanda kudzidziswa
   • Iyo probe inogona kunge iri kushandisa hukama hwenhema pane iyo pfungwa yechokwadi
   • Linear probes haigone kuwanda
   • Cross-entropy kurasikirwa haigone kushandiswa
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