GALAXY: Graph-based Active Learning at the Extreme

Zhang, Jifan; Katz-Samuels, Julian; Nowak, Robert

doi:10.48550/arxiv.2202.01402

Cited by 1 publication

(11 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By adopting this view, one can ran-domly interleave the above-mentioned AL algorithms for every class. In this paper, we include baselines derived from least confidence sampling (Settles, 2009), GALAXY (Zhang et al, 2022) and most likely positive sampling (Warmuth et al, 2001;2003;Jiang et al, 2018).…”

Section: Related Workmentioning

confidence: 99%

“…Unbalanced Multi-class Deep Active Learning. More general and prevalent scenarios, such as unbalanced deep active classification, have received increasing attention in recent years (Kothawade et al, 2021;Emam et al, 2021;Zhang et al, 2022;Coleman et al, 2022;Jin et al, 2022;Aggarwal et al, 2020;Cai, 2022). For instance, Kothawade et al (2021) label examples with gradient embeddings that are most similar to previously collected rare examples while most dissimilar to out-of-distribution ones.…”

Section: Related Workmentioning

confidence: 99%

“…For instance, Kothawade et al (2021) label examples with gradient embeddings that are most similar to previously collected rare examples while most dissimilar to out-of-distribution ones. Zhang et al (2022) create linear one-vs-rest graphs based on margin scores. To collect a more class-diverse labeled set, GALAXY discovers and labels around the optimal uncertainty thresholds through a bisection procedure on shortest shortest paths.…”

Section: Related Workmentioning

confidence: 99%

“…• To our knowledge, we propose the first adaptive algorithm selection strategy for deep active learning. Our algorithm TAILOR works particularly well on the challenging and prevalent class-imbalance settings (Kothawade et al, 2021;Emam et al, 2021;Zhang et al, 2022). • Our framework is general purpose for both multi-label and multi-class classification.…”

Section: Introductionmentioning

confidence: 97%

“…Label-efficiency is increasingly crucial as deep learning models require large amount of labeled training data. In recent years, numerous new algorithms have been proposed for deep active learning (Sener & Savarese, 2017;Gal et al, 2017;Ash et al, 2019;Kothawade et al, 2021;Citovsky et al, 2021;Zhang et al, 2022). Relative label efficiencies among algorithms, however, vary significantly across datasets and applications (Beck et al, 2021;Zhan et al, 2022).…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Deep Generative Model for Multi-Class Imbalanced Learning

Zhang¹

View full text Add to dashboard Cite

Label efficiency has become an increasingly important objective in deep learning applications. Active learning aims to reduce the number of labeled examples needed to train deep networks, but the empirical performance of active learning algorithms can vary dramatically across datasets and applications. It is difficult to know in advance which active learning strategy will perform well or best in a given application. To address this, we propose the first adaptive algorithm selection strategy for deep active learning. For any unlabeled dataset, our (meta) algorithm TAILOR (Thompson ActIve Learning algORithm selection) iteratively and adaptively chooses among a set of candidate active learning algorithms. TAILOR uses novel reward functions aimed at gathering class-balanced examples. Extensive experiments in multi-class and multi-label applications demonstrate TAILOR 's effectiveness in achieving accuracy comparable or better than that of the best of the candidate algorithms.

show abstract