| 2019 |
Single shot active learning using pseudo annotators |
Yang and Loog |
Patern Recognition |
- |
Challenge, single shot setting, standard active learning algorithms, New annotation method, CNNs |
human annotators should be always readily available for labeling whenever new unlabeled samples are queried. However, this assump- tion may not hold in some real-world applications since (1) human annotator is unlikely to be present at all time, e.g. human annotator may get tired or need a rest, (2) and ac- tive learning process has to be suspended until the annota- tor reappear. |
MNIST, USPS dataset, Amazon, Webcam and Caltech datasets |
| 2019 |
A novel active learning framework for classification: using weighted rank aggregation to achieve multiple query criteria |
Yang and Loog |
Patern Recognition |
- |
multiple query strategies, Challenge |
(1) To the best of our knowledge, this is the first work to analyze and induct the existing MQCAL method with different integration criteria strategies. (2) This is also the first work to implement the MQCAL method by introducing weighted rank aggregation approaches, and the proposed framework may inspire future AL. (3) We present a mechanism that allows for a dynamic and self-adaptive tradeoff between any number and kind of involved SQC in a unified system by introducing the BVSB strategy. (4) We summarize basic rules for the use of our RMQCAL. The potentially best combination of involved SQC and rank aggregation approaches is also found from experimental comparative results. (5) Several comparative experiments are conducted to prove the effectiveness of the proposed RMQCAL method in many public data sets |
binary classification problems through the UCI Repository |
| 2019 |
Inductive conformal predictor for convolutional neural networks: Applications to active learning for image classification |
Matiz and Barner |
Patern Recognition |
- |
Image Classification , Hybrid Method, Conformal prediction |
Conformal prediction uses the degree of strangeness (nonconformity) of data instances to determine the confidence values of new predictions. We propose an inductive conformal predictor for convolutional neural networks (CNNs), referring to it as ICP-CNN, which uses a novel nonconformity measure that produces reliable confidence values. Furthermore, ICP-CNN is used to improve classification performance through active learning, selecting instances from an unlabeled pool based on the evaluation of three cri- teria: informativeness, diversity, and information density. Distance metric learning is employed to mea- sure diversity, using a similarity measure that adapts to the database being used. Moreover, information density is considered to filter outliers. Experiments conducted on face and object recognition databases demonstrate that ICP-CNN improves the classification performance of CNNs, outperforming previously proposed active learning techniques, while producing reliable confidence values. |
YaleB database, AR Database, Caltech101｜ |
| 2019 |
A novel active learning framework for classification: Using weighted rank aggregation to achieve multiple query criteria |
Yu et al. |
Patern Recognition |
- |
Hybrid Selection, Classification Task, Person re-identification, CNNs, Increatment learning |
Annotating a large-scale image dataset is very tedious, yet necessary for training person re-identification (re-ID) models. To alleviate such a problem, we present an active redundancy reduction (ARR) framework via training an effective re-ID model with the least labeling efforts. The proposed ARR framework actively selects informative and diverse samples for annotation by estimating their uncertainty and intra-diversity, thus it can significantly reduce the annotation workload. Moreover, we propose a computer-assisted iden- tity recommendation module embedded in the ARR framework to help human annotators to rapidly and accurately label the selected samples. Extensive experiments were carried out on several public re-ID datasets to demonstrate the existence of data redundancy. Experimental results indicate that our method can reduce 57%, 63%, and 49% annotation efforts on the Market1501, MSMT17, and CUHK03, respectively, while maximizing the performance of the re-ID model. |
Market1501, MSMT17, and CUHK03 |