Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo Cyumurongo hamwe na Frozen Ikiranga Isuzuma Ikibazo

Kugenzura ibipimo byumurongo bipima, uburyo bwo kugereranya ibice byafunzwe nimpamvu iperereza ryonyine ryonyine ridashobora gushiraho impamvu cyangwa imbaraga.

Ikibazo 1 Bya 8

Muburyo busanzwe bwumurongo, ni ibihe bipimo bigezweho mugihe cy'amahugurwa?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Muburyo busanzwe bwumurongo, ni ibihe bipimo bigezweho mugihe cy'amahugurwa?
    • Gusa ibishya byongeweho umurongo utondekanya hejuru
    • Umugongo wose wongeyeho umurongo ugaragara
    • Gusa umugongo, usize urutonde rwakosowe
    • Igice cyo gushiramo hamwe nibitekerezo byanyuma
  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. Ugereranije nuburyo bwuzuye-buringaniza, umurongo ugenzura ushimishije cyane kuko ni:
    • Birenzeho kuri buri gikorwa
    • Bijejwe gushyira muri rusange ibyiza bitagabanijwe
    • Guhendutse muri compte na memoire kandi itanga isuku isomwe kumiterere yibiranga
    • Inzira yonyine yo gukoresha icyitegererezo cyateganijwe
  4. Ikintu gisanzwe gifatika mugihe cyo gusuzuma impinduka ni:
    • Igice cyanyuma burigihe gitanga ibyiza byukuri
    • Ikibazo gikora gusa kumurongo wemeza
    • Guhitamo ibice ntabwo bigira ingaruka kubisubizo byubushakashatsi
    • Hagati aho akenshi iruta urwego rwa nyuma rwo kwimura
  5. Kurenga ibipimo, umurongo ugereranije urakoreshwa mugusobanurwa kuri:
    • Menya niba igitekerezo kimeze nkukuri kiba muburyo bwihishe
    • Kanda moderi zoherejwe na mobile
    • Simbuza optimizer mugihe cyo kwitegura
    • Kora amakuru mashya mumahugurwa mu buryo bwikora
  6. Ibyingenzi byingenzi byuburyo bukoreshwa mugusobanura ukuri gukomeye ni uko:
    • Umugongo ukomeza rwihishwa imyitozo
    • Iperereza rishobora kuba rikoresha isano ridasanzwe aho kuba igitekerezo nyacyo
    • Imirongo igororotse ntishobora kurenza urugero
    • Igihombo cya cross-entropy ntigishobora gukoreshwa
  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