Batch mode active learning and its application to medical image classification

By Steven C. H. Hoi, Rong Jin, Jianke Zhu, Michael R. Lyu. 2006

521 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Learning Algorithms for Active Learning (2017, background): “Likewise, collecting labels for a medical imaging task may be costly because it requires a specialist (Hoi et al., 2006), and the cost could be reduced by carefully selecting which images to label.”
Near-optimal Observation Selection using Submodular Functions (2007, background): “Hoi et al. (2006) show that certain active learning objectives are (approximately) submodular, hence our algorithms could potentially be used in this context.”
Collaborative Unsupervised Domain Adaptation for Medical Image Diagnosis (2019, background): “since medical images can only be annotated by doctors with expertise [24].”
Restricted Strong Convexity Implies Weak Submodularity (2016, background): “In parallel work, several authors have demonstrated that the subset selection problem can be connected to submodular optimization [2, 20, 48, 49] and that greedy algorithms are widely used for iteratively building good feature sets.”
BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning (2019, background): “AL has already made real-world impact in manufacturing [34], robotics [5], recommender systems [1], medical imaging [18], and NLP [31], motivating the need for pushing AL even further.”
Discriminative Batch Mode Active Learning (2007, background): “Our empirical studies on UCI datasets show that the proposed active learning is more effective than current state-of-the art batch mode active learning algorithms.”
Active Learning for Convolutional Neural Networks: A Core-Set Approach (2017, background): “…also many pool based active learning algorithms designed for the specific class of machine learning algorithms like k-nearest neighbors and naive Bayes (Wei et al., 2015), logistic regression Hoi et al. (2006); Guo & Schuurmans (2008), and linear regression with Gaussian noise (Yu et al., 2006).”
3D Deep Learning on Medical Images: A Review (2020, methodology): “Historical methods for automated classification of images involves extensive rule-based algorithms or manual feature handcrafting [16–21], which are time-consuming, have poor generalization capacity and require domain knowledge.”
A Survey of Deep Active Learning (2020, background): “labeled datasets is relatively simple, the manual labeling of datasets comes at a high cost; this is particularly true for those fields where labeling requires a high degree of professional knowledge [64, 153]. For example, the labeling and description of lung lesion images of COVID-19 patients requ”
The Power of Ensembles for Active Learning in Image Classification (2018, background): “Commonly used acquisition functions are based on criteria such as uncertainty of prediction ([16], [32], [62]), coverage of the data space ([52]), unanimity of a committee ([30]), expected error, or variance reduction ([25], [33]).”
A survey on instance selection for active learning (2012, background): “[20] solves the Semi-Definitive Programming problem by exploring the properties of submodular function.”
Human-in-the-loop machine learning: a state of the art (2022, background): “Regarding image and video classification, in the medical context we can find AL in the classification of radiology reports (Hoi et al. 2006; Nguyen and Patrick 2014).”
Active Learning Literature Survey (2009, methodology): “It has been suggested that uncertainty sampling and QBC strategies are prone to querying outliers, which is a main motivating factor behind the Fisher information and estimated error reduction frameworks (Roy and McCallum, 2001; Zhu et al., 2003; Hoi et al., 2006b).”
Learning Active Learning from Data (2017, background): “They include uncertainty sampling [32, 15, 27, 34], query-by-committee [7, 13], expected model change [27, 30, 33], expected error or variance minimization [14, 9] and information gain [10].”
From Theories to Queries (2011, methodology): “While these approaches use greedy heuristics, Hoi et al. (2006b) exploit the properties of a submodular functions (Nemhauser et al., 1978) to find a batch that is guaranteed to be near-optimal.”