Scalable Batch-Mode Deep Bayesian Active Learning via Equivalence Class Annealing

Authors: Renyu Zhang, Aly A Khan, Robert L. Grossman, Yuxin Chen

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this paper, we propose Batch-BALANCE, a scalable batch-mode active learning algorithm... We show that our algorithm can effectively handle realistic multi-class classification tasks, and achieves compelling performance on several benchmark datasets for active learning under both lowand large-batch regimes.
Researcher Affiliation Academia Renyu Zhang1, Aly A. Khan2,3, Robert L. Grossman1,4, Yuxin Chen1 1Department of Computer Science, University of Chicago 2Department of Pathology, University of Chicago 3Department of Family Medicine, University of Chicago 4Department of Medicine, University of Chicago
Pseudocode Yes Algorithm 1 Active selection w/ Batch-BALANCE; Algorithm 2 Greedy Selection; Algorithm 3 BALANCE-Clustering
Open Source Code Yes Reference code is released at https://github.com/zhangrenyuuchicago/BALan Ce.
Open Datasets Yes In the main paper, we consider four datasets (i.e. MNIST (Le Cun et al., 1998), Repeated MNIST (Kirsch et al., 2019), Fashion-MNIST (Xiao et al., 2017) and EMNIST (Cohen et al., 2017)) as benchmarks for the small-batch setting, and two datasets (i.e. SVHN (Netzer et al., 2011), CIFAR (Krizhevsky et al., 2009)) as benchmarks for the large-batch setting.
Dataset Splits Yes We split each dataset into unlabeled AL pool Dpool, initial training dataset Dtrain, validation dataset Dval, test dataset Dtest and unlabeled dataset Dpool.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions using 'Adam optimizer' but does not specify any software libraries or frameworks (e.g., PyTorch, TensorFlow) with their version numbers.
Experiment Setup Yes To avoid overfitting, we train the BNNs with MC dropout at each iteration with early stopping. for MNIST, Repeated-MNIST, EMNIST, and Fashion MNIST, we terminate the training of BNNs with patience of 3 epochs. For SVHN and CIFAR-10, we terminate the training of BNNs with patience of 20 epochs. [...] We use Adam optimizer (Kingma & Ba, 2017) for all the models in the experiments. For c SG-MCMC, we use Res Net-18 (He et al., 2016) and run 400 epochs in each AL iteration. We set the number of cycles to 8 and initial step size to 0.5. [...] VGG-11 is trained using SGD with fixed learning rate 0.001 and momentum 0.9.