On Power Laws in Deep Ensembles

Authors: Ekaterina Lobacheva, Nadezhda Chirkova, Maxim Kodryan, Dmitry P. Vetrov

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct our experiments with convolutional neural networks, Wide Res Net [33] and VGG16 [27], on CIFAR-10 [16] and CIFAR-100 [17] datasets. [...] The empirical results presented in sections 4, 5, 6, 7 were supported by the Russian Science Foundation grant 19-71-30020.
Researcher Affiliation Collaboration 1Samsung-HSE Laboratory, National Research University Higher School of Economics 2Samsung AI Center Moscow Moscow, Russia {elobacheva,nchirkova,mkodryan,dvetrov}@hse.ru
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Our source code is available at https://github.com/nadiinchi/power_laws_deep_ensembles.
Open Datasets Yes We conduct our experiments with convolutional neural networks, Wide Res Net [33] and VGG16 [27], on CIFAR-10 [16] and CIFAR-100 [17] datasets.
Dataset Splits Yes Following [2], we use the test-time cross-validation to compute the CNLL. [...] For each network size, we tune hyperparameters (weight decay and dropout) using grid search.
Hardware Specification No The paper mentions support from 'HPC facilities at NRU HSE' but does not provide specific details on CPU, GPU models, or other hardware specifications used for experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For each network size, we tune hyperparameters (weight decay and dropout) using grid search. We train all networks for 200 epochs with SGD with an annealing learning schedule and a batch size of 128.