Semantic Scholar Open Access 2021 490 sitasi

Revisiting the Calibration of Modern Neural Networks

M. Minderer J. Djolonga Rob Romijnders F. Hubis Xiaohua Zhai +3 lainnya

Abstrak

Accurate estimation of predictive uncertainty (model calibration) is essential for the safe application of neural networks. Many instances of miscalibration in modern neural networks have been reported, suggesting a trend that newer, more accurate models produce poorly calibrated predictions. Here, we revisit this question for recent state-of-the-art image classification models. We systematically relate model calibration and accuracy, and find that the most recent models, notably those not using convolutions, are among the best calibrated. Trends observed in prior model generations, such as decay of calibration with distribution shift or model size, are less pronounced in recent architectures. We also show that model size and amount of pretraining do not fully explain these differences, suggesting that architecture is a major determinant of calibration properties.

Topik & Kata Kunci

Penulis (8)

M

M. Minderer

J

J. Djolonga

R

Rob Romijnders

F

F. Hubis

X

Xiaohua Zhai

N

N. Houlsby

D

Dustin Tran

M

M. Lučić

Format Sitasi

Minderer, M., Djolonga, J., Romijnders, R., Hubis, F., Zhai, X., Houlsby, N. et al. (2021). Revisiting the Calibration of Modern Neural Networks. https://www.semanticscholar.org/paper/e757488d2e8684e3da7b14fbb000b7e4a0bab001

Akses Cepat

PDF tidak tersedia langsung

Cek di sumber asli →
Lihat di Sumber
Informasi Jurnal
Tahun Terbit
2021
Bahasa
en
Total Sitasi
490×
Sumber Database
Semantic Scholar
Akses
Open Access ✓