Confidence-Aware Learning for Deep Neural Networks

Authors: Jooyoung Moon, Jihyo Kim, Younghak Shin, Sangheum Hwang

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results on classification benchmark datasets indicate that the proposed method helps networks to produce well-ranked confidence estimates.
Researcher Affiliation Collaboration 1Department of Data Science, Seoul National University of Science and Technology, Seoul, Republic of Korea 2LG CNS, Seoul, Republic of Korea 3Department of Industrial & Information Systems Engineering, Seoul National University of Science and Technology, Seoul, Republic of Korea.
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/daintlab/confidenceaware-learning.
Open Datasets Yes We evaluate our method on benchmark datasets for image classification: SVHN (Netzer et al., 2011) and CIFAR-10/100 (Krizhevsky & Hinton, 2009).
Dataset Splits No The paper mentions training and testing on benchmark datasets (SVHN, CIFAR-10/100) which have standard splits, but it does not explicitly provide specific percentages, sample counts, or a detailed methodology for a distinct validation split beyond the standard training data used.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments.
Software Dependencies No The paper mentions software components and frameworks (e.g., SGD, MCdropout, PyTorch via a GitHub link for a model architecture) but does not provide specific version numbers for any ancillary software dependencies.
Experiment Setup Yes All models are trained using SGD with a momentum of 0.9, an initial learning rate of 0.1, and a weight decay of 0.0001 for 300 epochs with the mini-batch size of 128. The learning rate is reduced by a factor of 10 at 150 epochs and 250 epochs.