Feature Space Particle Inference for Neural Network Ensembles

Authors: Shingo Yashima, Teppei Suzuki, Kohta Ishikawa, Ikuro Sato, Rei Kawakami

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive evaluation on real-world datasets shows that our model significantly outperforms the gold-standard Deep Ensembles on various metrics, including accuracy, calibration, and robustness. and 4. Experiments In this section, we present the results on popular image classification tasks: CIFAR-10, CIFAR-100 (Krizhevsky & Hinton, 2009), and Image Net (Deng et al., 2009).
Researcher Affiliation Collaboration 1Denso IT Laboratory Inc., Tokyo, Japan 2Tokyo Institute of Technology, Tokyo, Japan.
Pseudocode Yes Algorithm 1 feature-WGD (in parallel)
Open Source Code Yes Code is available at: https: //github.com/Denso ITLab/feature PI.
Open Datasets Yes In this section, we present the results on popular image classification tasks: CIFAR-10, CIFAR-100 (Krizhevsky & Hinton, 2009), and Image Net (Deng et al., 2009).
Dataset Splits No The paper mentions using 'standard scheduling, augmentation, and regularization schemes in the literature' and references datasets like CIFAR and ImageNet, which typically have predefined splits. However, it does not explicitly state the specific percentages or counts for training, validation, and test splits used in their experiments.
Hardware Specification Yes In our experiment on CIFAR-100 for Wide Res Net-16-4 with an ensemble size of 10, feature-WGD takes approximately 17s for one epoch, while Deep Ensembles take 16s on four A100 GPUs.
Software Dependencies No The paper mentions using 'stochastic gradient descent with Nesterov momentum for optimization' and 'Adam' but does not specify software names with version numbers for libraries, frameworks, or programming languages (e.g., PyTorch version, Python version, CUDA version).
Experiment Setup Yes Table 7. Hyperparameter values for training on CIFAR-10, CIFAR-100, and Image Net. DATASET CIFAR-10 CIFAR-100 IMAGENET EPOCH 300 300 90 BATCH SIZE 128 128 256 BASE LEARNING RATE 0.1 0.1 0.1 LR DECAY RATIO 0.1 0.1 0.1 LR DECAY EPOCHS [150, 225] [150, 225, 250] [30, 60] MOMEMTUM 0.9 0.9 0.9 WEIGHT DECAY 5 10 4 5 10 4 1 10 4