Balanced Active Learning Method for Image Classification

Papp, Dávid; Szücs, Gábor

doi:10.14232/actacyb.23.2.2017.13

Cited by 7 publications

(4 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, to evaluate the first step we used sensitivity (TP rate) and specificity (TN rate). Moreover, we used the accuracy (Equation 39) and the balanced accuracy 1 (Equation 40) to measure the performance of the overall classification task [35], because unbalanced data might mislead the conclusion [29]. Lastly, we used mean squared error for evaluating the goodness of Formula 2 using Formula 37 (∆A K+1 ), and aggregating these results of all the experiments.…”

Section: Metricsmentioning

confidence: 99%

Single and Combined Algorithms for Open Set Classification on Image Datasets

Al-Shouha,

Szűcs

2024

Acta Cybern

Self Cite

View full text Add to dashboard Cite

Generally, classification models have closed nature, and they are constrained by the number of classes in the training data. Hence, classifying "unknown" - OOD (out-of-distribution) - samples is challenging, especially in the so called "open set" problem. We propose and investigate different solutions - single and combined algorithms - to tackle this task, where we use and expand a K-classifier to be able to identify K+1 classes. They do not require any retraining or modification on the K-classifier architecture. We show their strengths when avoiding type I or type II errors is fundamental. We also present a mathematical representation for the task to estimate the K+1 classification accuracy, and an inequality that defines its boundaries. Additionally, we introduce a formula to calculate the exact K+1 classification accuracy.

show abstract

Section: Metricsmentioning

confidence: 99%

Single and Combined Algorithms for Open Set Classification on Image Datasets

Al-Shouha,

Szűcs

2024

Acta Cybern

Self Cite

View full text Add to dashboard Cite

show abstract

“…(ŷ |̂, ) = ∑ (̂, ) =1 (14) where , are the samples, and their associated labels from the = {( , )} =1 support set, and (•,•) is the kernel (also known as attention kernel or attention mechanism). It is worth noting that the above relation produces the output (label) of the samples of the new classes as a linear combination of the sample labels in the support set.…”

Section: Figure 1 Matching Network Architecturementioning

confidence: 99%

“…This scenario is a particularly hot issue nowadays: how could a new disease for which only limited data are available be diagnosed using features of previous diseases? (If the number of labeled data is small, but the huge amount of unlabeled data is available, then this can lead to active learning [14], but in this paper, we consider that there is no unlabeled data at all. )…”

Section: Introductionmentioning

confidence: 99%

Double-View Matching Network for Few-Shot Learning to Classify Covid-19 in X-ray images

Szücs¹,

Nemeth²

2021

Infocommunications journal

Self Cite

View full text Add to dashboard Cite

The research topic presented in this paper belongs to small training data problem in machine learning (especially in deep learning), it intends to help the work of those working in medicine by analyzing pathological X-ray recordings, using only very few images. This scenario is a particularly hot issue nowadays: how could a new disease for which only limited data are available be diagnosed using features of previous diseases? In this problem, so-called few-shot learning, the difficulty of the classification task is to learn the unique feature characteristics associated with the classes. Although there are solutions, but if the images come from different views, they will not handle these views well. We proposed an improved method, so-called Double-View Matching Network (DVMN based on the deep neural network), which solves the few-shot learning problem as well as the different views of the pathological recordings in the images. The main contribution of this is the convolutional neural network for feature extraction and handling the multi-view in image representation. Our method was tested in the classification of images showing unknown COVID-19 symptoms in an environment designed for learning a few samples, with prior meta-learning on images of other diseases only. The results show that DVMN reaches better accuracy on multi-view dataset than simple Matching Network without multi-view handling.

show abstract

“…uncertainty sampling [6], query-by-committee (QBC) [25], expected model change [2], expected error reduction [14], or density-weighted method [1]. On the other hand, there are recently proposed query strategies, like uncertainty sampling with diversity maximization [29], Balanced Active Learning (BAL) method [17], extended margin and soft balanced strategy [18], Prototype Based Active Learning (PBAC) algorithm [3] and the hybrid, Expected Difference Change (EDC) [19]. However, these approaches expect the L to be not empty, because all of them applies some kind of supervised machine learning algorithm (e.g.…”

Section: Related Workmentioning

confidence: 99%

Zero Initialized Active Learning with Spectral Clustering using Hungarian Method

Papp¹

2021

Acta Cybern

Self Cite

View full text Add to dashboard Cite

Supervised machine learning tasks often require a large number of labeled training data to set up a model, and then prediction - for example the classification - is carried out based on this model. Nowadays tremendous amount of data is available on the web or in data warehouses, although only a portion of those data is annotated and the labeling process can be tedious, expensive and time consuming. Active learning tries to overcome this problem by reducing the labeling cost through allowing the learning system to iteratively select the data from which it learns. In special case of active learning, the process starts from zero initialized scenario, where the labeled training dataset is empty, and therefore only unsupervised methods can be performed. In this paper a novel query strategy framework is presented for this problem, called Clustering Based Balanced Sampling Framework (CBBSF), which is not only select the initial labeled training dataset, but uniformly selects the items among the categories to get a balanced labeled training dataset. The framework includes an assignment technique to implicitly determine the class membership probabilities. Assignment solution is updated during CBBSF iterations, hence it simulates supervised machine learning more accurately as the process progresses. The proposed Spectral Clustering Based Sampling (SCBS) query startegy realizes the CBBSF framework, and therefore it is applicable in the special zero initialized situation. This selection approach uses ClusterGAN (Clustering using Generative Adversarial Networks) integrated in the spectral clustering algorithm and then it selects an unlabeled instance depending on the class membership probabilities. Global and local versions of SCBS were developed, furthermore, most confident and minimal entropy measures were calculated, thus four different SCBS variants were examined in total. Experimental evaluation was conducted on the MNIST dataset, and the results showed that SCBS outperforms the state-of-the-art zero initialized active learning query strategies.

show abstract

Balanced Active Learning Method for Image Classification

Cited by 7 publications

References 20 publications

Single and Combined Algorithms for Open Set Classification on Image Datasets

Single and Combined Algorithms for Open Set Classification on Image Datasets

Double-View Matching Network for Few-Shot Learning to Classify Covid-19 in X-ray images

Zero Initialized Active Learning with Spectral Clustering using Hungarian Method

Contact Info

Product

Resources

About