Active Learning in Bayesian Neural Networks with Balanced Entropy Learning Principle

Authors: Jae Oh Woo

ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we demonstrate that our balanced entropy learning principle with Bal Ent Acq1 consistently outperforms well-known linearly scalable active learning methods, including a recently proposed Power BALD, a simple but diversified version of BALD, by showing experimental results obtained from MNIST, CIFAR-100, SVHN, and Tiny Image Net datasets.
Researcher Affiliation Collaboration Jae Oh Woo Samsung SDS Research America San Jose, CA 95134 jaeoh.woo@aya.yale.edu
Pseudocode Yes Algorithm 1: Bal Ent Acq active learning algorithm
Open Source Code Yes 1Code is available. https://github.com/jaeohwoo/Balanced Entropy
Open Datasets Yes obtained from MNIST (Le Cun & Cortes, 2010), CIFAR-100 (Krizhevsky et al., 2012), SVHN (Netzer et al., 2011), and Tiny Image Net (Le & Yang, 2015) datasets
Dataset Splits No The paper specifies 'Train Size' and 'Test size' in tables (e.g., Table 1, Table 2) but does not explicitly mention or detail a separate validation set split or how validation was used in the experimental setup.
Hardware Specification Yes We used a single NVIDIA A100 GPU for each experiment
Software Dependencies No The paper mentions software components implicitly through common frameworks (e.g., Adam optimizer, ResNet architectures) but does not provide specific version numbers for any libraries or software dependencies like Python, PyTorch, or CUDA.
Experiment Setup Yes Table 2 shows a summary of dataset, configurations, and hyperparmeters used in our experiments. For each experiment, we repeat 3 times to generate the full active learning accuracy curve.