On Calibration of Modern Neural Networks
An influential 2017 paper by Guo et al. in the field of deep learning that systematically documents the overconfidence problem in modern deep neural networks and proposes calibration methods including temperature scaling.
Timeline (last 90 days)
The 2017 paper "On Calibration of Modern Neural Networks" by Guo et al. systematically revealed that increased model depth, use of BatchNorm layers, and longer training periods all exacerbate miscalibration.
The 2017 paper "On Calibration of Modern Neural Networks" by Guo et al. systematically revealed the problem of poor calibration, finding that increased model depth, the use of BatchNorm layers, and longer training periods all exacerbate poor calibration.
All Facts (2)
The 2017 paper "On Calibration of Modern Neural Networks" by Guo et al. systematically revealed that increased model depth, use of BatchNorm layers, and longer training periods all exacerbate miscalibration.
50%UnverifiedThe 2017 paper "On Calibration of Modern Neural Networks" by Guo et al. systematically revealed the problem of poor calibration, finding that increased model depth, the use of BatchNorm layers, and longer training periods all exacerbate poor calibration.
50%