Ikibazo gihuza ibibazo · Hagati Urwego
Ikibazo cyo Guhindura Ikibazo
Kugenzura ibisobanuro bya kalibrasi, igishushanyo cyizewe, ECE, igipimo cy'ubushyuhe n'impamvu ireme ryemewe ritandukanijwe nukuri.
Ikibazo 1 Bya 8
Icyitegererezo kirahinduwe neza niba, mubyahanuwe biranga 80% byizeye, ibyabaye bibaho:
Reba buri kibazo8 ibibazo
Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.
- Icyitegererezo kirahinduwe neza niba, mubyahanuwe biranga 80% byizeye, ibyabaye bibaho:
- Hafi ya 80% yigihe
- 100% yigihe
- Hafi 20% yigihe
- Rimwe
- Igipimo cy'ubushyuhe gikora iki ku miyoboro y'urusobe rw'imitsi?
- Guhindura icyiciro cyahanuwe
- Kugabanya logi na scalar wize mbere ya softmax, uhindura ikizere
- Ongeraho urusaku kubirango byamahugurwa
- Kongera umuyoboro kuva kera
- Ni ikihe gishushanyo mbonera cyizewe?
- Gutakaza imyitozo n'ibihe
- Ingano yicyitegererezo nukuri
- Biteganijwe ko bishoboka hamwe ninshuro zagaragaye
- Igipimo cyo kwiga nubunini bwicyiciro
- Biteganijwe ko Calibration Error (ECE) ingamba:
- Icyitegererezo cyo hejuru-1 neza
- Umubare wibipimo murugero
- Itandukaniro rya gradients
- Ikigereranyo kiremereye hagati yicyizere nukuri kuri bine
- Imiyoboro yimbitse igezweho ikunze gusobanurwa nk:
- Kurenza urugero
- Kutizera
- Guhinduranya neza muburyo busanzwe
- Ntibishoboka gusohora ibishoboka
- Ni ukubera iki icyitegererezo cyukuri gishobora gukenera kalibrasi?
- Moderi yukuri ihora ifite ikosa rya zeru
- Ukuri na kalibrasi nibintu bitandukanye
- Calibration ifite akamaro gusa ko gusubira inyuma
- Ukuri kuremeza ko bishoboka
- On which data is temperature scaling typically fit in the guide’s calibration workflow?
- The final test set using classification accuracy
- Only the training labels without model outputs
- A held-out validation set using negative log-likelihood
- The model’s hidden weights using mean squared error
- Why can temperature scaling improve calibration without changing top-1 accuracy?
- It changes the model’s predicted class labels randomly
- It retrains the feature extractor
- It replaces every logit with the same value
- Dividing logits by a positive scalar changes confidence but preserves their ordering