Wyniki wyszukiwania - BazTech

1

Towards explainable classifiers using the counterfactual approach : global explanations for discovering bias in data

Mikołajczyk Agnieszka, Grochowski Michał, Kwasigroch Arkadiusz

Journal of Artificial Intelligence and Soft Computing Research

|

2021

|

Vol. 11, No. 1

51--67

EN

The paper proposes summarized attribution-based post-hoc explanations for the detection and identification of bias in data. A global explanation is proposed, and a step-by-step framework on how to detect and test bias is introduced. Since removing unwanted bias is often a complicated and tremendous task, it is automatically inserted, instead. Then, the bias is evaluated with the proposed counterfactual approach. The obtained results are validated on a sample skin lesion dataset. Using the proposed method, a number of possible bias-causing artifacts are successfully identified and confirmed in dermoscopy images. In particular, it is confirmed that black frames have a strong influence on Convolutional Neural Network’s prediction: 22% of them changed the prediction from benign to malignant.

2

Analiza sceny przy użyciu głębokich sieci neuronowych typu YOLO

Mikołajczyk Mateusz, Kwasigroch Arkadiusz, Grochowski Michał

Zeszyty Naukowe Wydziału Elektrotechniki i Automatyki Politechniki Gdańskiej

|

2019

|

Nr 68

37--40

PL

W artykule opisany został problem analizy sceny na obrazach oraz sekwencjach video. Zadanie analizy sceny polega na detekcji, lokalizacji i klasyfikacji obiektów znajdujących się na obrazach. Zaimplementowany system wykorzystuje głęboką sieć neuronową, której struktura oparta została na architekturze YOLO (You Only Look Once). Niskie zapotrzebowania obliczeniowe wybranej architektury pozwala na wykonywanie detekcji w czasie rzeczywistym z zadowalającą dokładnością. W pracy przeprowadzono również badania nad doborem odpowiedniego optymalizatora wykorzystywanego w procesie uczenia. Jako przykładową aplikację wybrano analizę ruchu ulicznego w której skład wchodzi wykrywanie i lokalizacja obiektów takich jak m.in. samochody, motocykle czy sygnalizacja świetlna. Systemy tego typu mogą być wykorzystywane w wszelkiego typu systemach analizy wizyjnej otoczenia np. w pojazdach autonomicznych, systemach automatycznej analizy video z kamer przemysłowych, systemach dozoru czy analizy zdjęć satelitarnych.

EN

The paper describes the problem of scene analysis in images and video sequences. The task of scene analysis is to detect, locate and classify objects in images. As an example of an application, traffic analysis was chosen, which includes the detection and location of objects such as cars, motorcycles or traffic lights. The implemented system uses a deep neural network, whose structure is based on the YOLO (You Only Look Once) architecture. Low computing requirements of the chosen architecture allows performing real-time detection with satisfactory accuracy. The work also included a study on the selection of an appropriate optimizer used in the learning process. The program correctly detects objects with a large surface area, allowing the system to be used in traffic analysis. The work also showed that using the ADAM algorithm allowed significantly shorten the training time of the neural network. Systems of this type can be used in many types of video analysis systems such as autonomous vehicles, automatic video analysis systems with CCTV cameras, surveillance systems or satellite image analysis.

3

Machine learning system for automated blood smear analysis

Grochowski Michał, Wąsowicz Michał, Mikołajczyk Agnieszka, Ficek Mateusz, Kulka Marek, Wróbel Maciej S., Jędrzejewska-Szczerska Małgorzata

Metrology and Measurement Systems

|

2019

|

Vol. 26, nr 1

81--93

EN

In this paper the authors propose a decision support system for automatic blood smear analysis based onmicroscopic images. The images are pre-processed in order to remove irrelevant elements and to enhancethe most important ones – the healthy blood cells (erythrocytes) and the pathologic ones (echinocytes). The separated blood cells are analysed in terms of their most important features by the eigenfaces method. The features are the basis for designing the neural network classifier, learned to distinguish between erythrocytes and echinocytes. As the result, the proposed system is able to analyse the smear blood images in a fully automatic way and to deliver information on the number and statistics of the red blood cells, both healthy and pathologic. The system was examined in two case studies, involving the canine and human blood, and then consulted with the experienced medicine specialists. The accuracy of classification of red blood cells into erythrocytes and echinocytes reaches 96%.

4

Diagnosis of malignant melanoma by neural network ensemble-based system utilising hand-crafted skin lesion features

Grochowski Michał, Mikołajczyk Agnieszka, Kwasigroch Arkadiusz

Metrology and Measurement Systems

|

2019

|

Vol. 26, nr 1

65--80

EN

Malignant melanomas are the most deadly type of skin cancer, yet detected early have high chances of successful treatment. In the last twenty years, the interest in automatic recognition and classification of melanoma dynamically increased, partly because of appearing public datasets with dermatoscopic images of skin lesions. Automated computer-aided skin cancer detection in dermatoscopic images is a very challenging task due to uneven sizes of datasets, huge intra-class variation with small interclass variation, and the existence of many artifacts in the images. One of the most recognized methods of melanoma diagnosis is the ABCD method. In the paper, we propose an extended version of this method and an intelligent decision support system based on neural networks that uses its results in the form of hand-crafted features. Automatic determination of the skin features with the ABCD method is difficult due to the large diversity of images of various quality, the existence of hair, different markers and other obstacles. Therefore, it was necessary to apply advanced methods of pre-processing the images. The proposed system is an ensemble of ten neural networks working in parallel, and one network using their results to generate a final decision. This system structure enables to increase the efficiency of its operation by several percentage points compared with asingle neural network. The proposed system is trained on over 5000 and tested afterwards on 200 skin moles. The presented system can be used as a decision support system for primary care physicians, as a system capable of self-examination of the skin with a dermatoscope and also as an important tool to improve biopsy decision making.

5

Selected technical issues of deep neural networks for image classification purposes

Grochowski Michał, Kwasigroch A., Mikołajczyk A.

Bulletin of the Polish Academy of Sciences. Technical Sciences

|

2019

|

Vol. 67, nr 2

363--376

EN

In recent years, deep learning and especially deep neural networks (DNN) have obtained amazing performance on a variety of problems, in particular in classification or pattern recognition. Among many kinds of DNNs, the convolutional neural networks (CNN) are most commonly used. However, due to their complexity, there are many problems related but not limited to optimizing network parameters, avoiding overfitting and ensuring good generalization abilities. Therefore, a number of methods have been proposed by the researchers to deal with these problems. In this paper, we present the results of applying different, recently developed methods to improve deep neural network training and operating. We decided to focus on the most popular CNN structures, namely on VGG based neural networks: VGG16, VGG11 and proposed by us VGG8. The tests were conducted on a real and very important problem of skin cancer detection. A publicly available dataset of skin lesions was used as a benchmark. We analyzed the influence of applying: dropout, batch normalization, model ensembling, and transfer learning. Moreover, the influence of the type of activation function was checked. In order to increase the objectivity of the results, each of the tested models was trained 6 times and their results were averaged. In addition, in order to mitigate the impact of the selection of learning, test and validation sets, k-fold validation was applied.