Maswali ya Tathmini ya Vipengee Vilivyogandishwa na Linear

Majibu na maelezo huonekana unapojibu maswali.

1. Katika uchunguzi wa kawaida wa mstari, ni vigezo gani vinasasishwa wakati wa mafunzo?
   • Kiainishi kipya cha mstari pekee kilichoongezwa hapo juu
   • Uti wa mgongo mzima pamoja na safu ya mstari
   • Uti wa mgongo tu, ukiacha kiainishaji kikiwa kimerekebishwa
   • Safu ya kupachika na kizuizi cha mwisho cha tahadhari
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. Ikilinganishwa na usanifu kamili, uchunguzi wa mstari unavutia kwa sababu ni:
   • Sahihi zaidi kwa kila kazi
   • Imehakikishwa kujumlisha vizuri zaidi nje ya usambazaji
   • Nafuu katika kuhesabu na kumbukumbu na inatoa usomaji safi zaidi juu ya ubora wa kipengele
   • Njia pekee ya kutumia mfano uliofundishwa
4. Ugunduzi wa kawaida wa nguvu wakati wa kuchunguza transfoma ni kwamba:
   • Safu ya mwisho daima inatoa usahihi bora wa uchunguzi
   • Uchunguzi hufanya kazi tu kwenye mitandao ya ushawishi
   • Chaguo la tabaka kamwe haliathiri matokeo ya uchunguzi
   • Tabaka za kati mara nyingi hushinda safu ya mwisho ya uhamishaji
5. Zaidi ya uwekaji alama, uchunguzi wa mstari unazidi kutumika katika kutafsiri kwa:
   • Tambua kama dhana kama ukweli imesimbwa katika hali fiche
   • Finya mifano kwa ajili ya kupelekwa kwa simu
   • Badilisha kiboreshaji wakati wa mazoezi ya awali
   • Tengeneza data mpya ya mafunzo kiotomatiki
6. Hatari kuu ya mbinu wakati wa kutafsiri usahihi wa uchunguzi wa juu ni kwamba:
   • Mgongo huweka mafunzo kwa siri
   • Uchunguzi unaweza kuwa unatumia uunganisho wa uwongo badala ya dhana ya kweli
   • Uchunguzi wa mstari hauwezi kupita kiasi
   • Hasara ya msalaba-entropy haiwezi kutumika
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