Search

Self-supervised learning has proven to be an effective way to learn representations in domains where annotated labels are scarce, such as medical imaging. A widely adopted framework for this purpose is contrastive learning and it has been applied to different scenarios. This paper seeks to advance our understanding of the contrastive learning framework by exploring a novel perspective: employing multi-organ datasets for pre-training models tailored to specific organ-related target tasks. More specifically, our target task is breast tumour segmentation in ultrasound images. The pre-training datasets include ultrasound images from other organs, such as the lungs and heart, and large datasets of natural images. Our results show that conventional contrastive learning pre-training improves performance compared to supervised baseline approaches. Furthermore, our pre-trained models achieve comparable performance when fine-tuned with only half of the available labelled data. Our findings also show the advantages of pre-training on diverse organ data for improving performance in the downstream task.

Electrocardiogram (ECG) signals, which capture the heart's electrical activity, are used to diagnose and monitor cardiac problems. The accurate classification of ECG signals, particularly for distinguishing among various types of arrhythmias and myocardial infarctions, is crucial for the early detection and treatment of heart-related diseases. This paper proposes a novel approach based on an improved differential evolution (DE) algorithm for ECG signal classification for enhancing the performance. In the initial stages of our approach, the preprocessing step is followed by the extraction of several significant features from the ECG signals. These extracted features are then provided as inputs to an enhanced multi-layer perceptron (MLP). While MLPs are still widely used for ECG signal classification, using gradient-based training methods, the most widely used algorithm for the training process, has significant disadvantages, such as the possibility of being stuck in local optimums. This paper employs an enhanced differential evolution (DE) algorithm for the training process as one of the most effective population-based algorithms. To this end, we improved DE based on a clustering-based strategy, opposition-based learning, and a local search. Clustering-based strategies can act as crossover operators, while the goal of the opposition operator is to improve the exploration of the DE algorithm. The weights and biases found by the improved DE algorithm are then fed into six gradient-based local search algorithms. In other words, the weights found by the DE are employed as an initialization point. Therefore, we introduced six different algorithms for the training process (in terms of different local search algorithms). In an extensive set of experiments, we showed that our proposed training algorithm could provide better results than the conventional training algorithms., Comment: 44 pages, 9 figures

The network layer is central to the networking scientific area. It is around the network layer that all the data communications develop, and one of its main tasks is to allow the identification of each single interface/machine between the potentially many interfaces in a network. This seminar addresses some of the issues that are usually presented to young Computer Science Engineering students in the course of several classes, but also presents some topics that are not address in networking courses. It is mostly focused on using Internet Protocol addresses in Local Area Networks, also considering issues that belong to the Wide Area Networks, such as data aggregation. This document summarizes the content of a seminar, therefore it comprehends both teaching and researching subject. The seminar starts with a history of the evolution of the communication protocols from the early days of networks up until IPv6. It describes a new approach to define the addresses of network interfaces using Variable Length Subnet Masks, as usually this is a not an easy task for Computer Science Engineering undergraduate students. This summary also describes some of the limitations of the data communication in todays' networks, proposing some solutions, where possible, including a novel mean of connectionless data transmission by using IPv6 addresses, by extension of previously published research. The way the seminar is organized provides a history to the past of the Internet Protocol, a view of some of its well-known current limitations, and a glimpse into a possible future regarding an improved connectionless layer 3 data transfer protocol.

The tremendous applications of human activity recognition are surging its span from health monitoring systems to virtual reality applications. Thus, the automatic recognition of daily life activities has become significant for numerous applications. In recent years, many datasets have been proposed to train the machine learning models for efficient monitoring and recognition of human daily living activities. However, the performance of machine learning models in activity recognition is crucially affected when there are incomplete activities in a dataset, i.e., having missing samples in dataset captures. Therefore, in this work, we propose a methodology for extrapolating the missing samples of a dataset to better recognize the human daily living activities. The proposed method efficiently pre-processes the data captures and utilizes the k-Nearest Neighbors (KNN) imputation technique to extrapolate the missing samples in dataset captures. The proposed methodology elegantly extrapolated a similar pattern of activities as they were in the real dataset., Comment: 8 Pages, 8 Figures, 1 Table. Accepted in 14th Multi Conference on Computer Science and Information Systems 2020 (MCCSIS 2020)

The User Datagram Protocol (UDP) and other similar protocols send the application data from the source machine to the destination machine inside segments, without foreseeing nor allowing for any type of control on the transmission or success metrics. These protocols are very convenient for e.g. real time data transmission. But when the reliability of the transmitted data is critical, other protocols termed as connection-oriented, allow for full control of the data transmission process, assuring that the received data is an exact copy of the transmitted data, e.g. the case of the Transmission Control Protocol (TCP). To sustain the increased functionality and features of the connection-oriented protocol, a set of mechanisms is implemented based on some specific fields of the segment header. These mechanisms result in a significant overhead in terms of the increased number of transmitted packets. This may further translate into significant delays, because of the additional number of switching and routing tasks, and eventually, because of more complex communications procedures, such as e.g. transmission window resizing, and of course, acknowledgement and sequence numbers updating. The two extremes of these communication modalities, one that has no control at all, and the other one that allows for full control, have resulted in the creation of an intermediate protocol that allows for a limited degree of knowledge on how successful a transmission was, and even for an eventual reordering of the segments that arrive out of sequence. This paper presents simulation results that confirm the efficiency of the new almost-reliable UDP protocol, termed Keyed User Datagram Protocol (or KUDP) for transmission of data that includes the ability to identify which packets were lost and to reorder packets that were received out-of-sequence, and points future tasks to be pursued in this research., Comment: 5 pages, 4 figures

Computer aided diagnosis (CAD) of Breast Cancer (BRCA) images has been an active area of research in recent years. The main goals of this research is to develop reliable automatic methods for detecting and diagnosing different types of BRCA from diagnostic images. In this paper, we present a review of the state of the art CAD methods applied to magnetic resonance (MRI) and mammography images of BRCA patients. The review aims to provide an extensive introduction to different features extracted from BRCA images through texture and statistical analysis and to categorize deep learning frameworks and data structures capable of using metadata to aggregate relevant information to assist oncologists and radiologists. We divide the existing literature according to the imaging modality and into radiomics, machine learning, or combination of both. We also emphasize the difference between each modality and methods strengths and weaknesses and analyze their performance in detecting BRCA through a quantitative comparison. We compare the results of various approaches for implementing CAD systems for the detection of BRCA. Each approachs standard workflow components are reviewed and summary tables provided. We present an extensive literature review of radiomics feature extraction techniques and machine learning methods applied in BRCA diagnosis and detection, focusing on data preparation, data structures, pre processing and post processing strategies available in the literature. There is a growing interest on radiomic feature extraction and machine learning methods for BRCA detection through histopathological images, MRI and mammography images. However, there isnt a CAD method able to combine distinct data types to provide the best diagnostic results. Employing data fusion techniques to medical images and patient data could lead to improved detection and classification results., Comment: 25 pages, 7 figures, Scientific grant report

In this work, we address the problem of learning an ensemble of specialist networks using multimodal data, while considering the realistic and challenging scenario of possible missing modalities at test time. Our goal is to leverage the complementary information of multiple modalities to the benefit of the ensemble and each individual network. We introduce a novel Distillation Multiple Choice Learning framework for multimodal data, where different modality networks learn in a cooperative setting from scratch, strengthening one another. The modality networks learned using our method achieve significantly higher accuracy than if trained separately, due to the guidance of other modalities. We evaluate this approach on three video action recognition benchmark datasets. We obtain state-of-the-art results in comparison to other approaches that work with missing modalities at test time.

Falling is a commonly occurring mishap with elderly people, which may cause serious injuries. Thus, rapid fall detection is very important in order to mitigate the severe effects of fall among the elderly people. Many fall monitoring systems based on the accelerometer have been proposed for the fall detection. However, many of them mistakenly identify the daily life activities as fall or fall as daily life activity. To this aim, an efficient machine learning-based fall detection algorithm has been proposed in this paper. The proposed algorithm detects fall with efficient sensitivity, specificity, and accuracy as compared to the state-of-the-art techniques. A publicly available dataset with a very simple and computationally efficient set of features is used to accurately detect the fall incident. The proposed algorithm reports and accuracy of 99.98% with the Support Vector Machine(SVM) classifier., Comment: 6 pages, SENSORDEVICES 2018, the Ninth International Conference on Sensor Device Technologies and Applications, Venice, Italy, 16-20 September 2018

Heterogeneous data modalities can provide complementary cues for several tasks, usually leading to more robust algorithms and better performance. However, while training data can be accurately collected to include a variety of sensory modalities, it is often the case that not all of them are available in real life (testing) scenarios, where a model has to be deployed. This raises the challenge of how to extract information from multimodal data in the training stage, in a form that can be exploited at test time, considering limitations such as noisy or missing modalities. This paper presents a new approach in this direction for RGB-D vision tasks, developed within the adversarial learning and privileged information frameworks. We consider the practical case of learning representations from depth and RGB videos, while relying only on RGB data at test time. We propose a new approach to train a hallucination network that learns to distill depth information via adversarial learning, resulting in a clean approach without several losses to balance or hyperparameters. We report state-of-the-art results on object classification on the NYUD dataset and video action recognition on the largest multimodal dataset available for this task, the NTU RGB+D, as well as on the Northwestern-UCLA., Comment: Accepted to IEEE Transactions on Pattern Analysis and Machine Intelligence

Diverse input data modalities can provide complementary cues for several tasks, usually leading to more robust algorithms and better performance. However, while a (training) dataset could be accurately designed to include a variety of sensory inputs, it is often the case that not all modalities could be available in real life (testing) scenarios, where a model has to be deployed. This raises the challenge of how to learn robust representations leveraging multimodal data in the training stage, while considering limitations at test time, such as noisy or missing modalities. This paper presents a new approach for multimodal video action recognition, developed within the unified frameworks of distillation and privileged information, named generalized distillation. Particularly, we consider the case of learning representations from depth and RGB videos, while relying on RGB data only at test time. We propose a new approach to train an hallucination network that learns to distill depth features through multiplicative connections of spatiotemporal representations, leveraging soft labels and hard labels, as well as distance between feature maps. We report state-of-the-art results on video action classification on the largest multimodal dataset available for this task, the NTU RGB+D. Code available at https://github.com/ncgarcia/modality-distillation ., Comment: Accepted at ECCV 2018; Supp. material at p.16; code available

The detection of the environment where user is located, is of extreme use for the identification of Activities of Daily Living (ADL). ADL can be identified by use of the sensors available in many off-the-shelf mobile devices, including magnetic and motion, and the environment can be also identified using acoustic sensors. The study presented in this paper is divided in two parts: firstly, we discuss the recognition of the environment using acoustic sensors (i.e., microphone), and secondly, we fuse this information with motion and magnetic sensors (i.e., motion and magnetic sensors) for the recognition of standing activities of daily living. The recognition of the environments and the ADL are performed using pattern recognition techniques, in order to develop a system that includes data acquisition, data processing, data fusion, and artificial intelligence methods. The artificial intelligence methods explored in this study are composed by different types of Artificial Neural Networks (ANN), comparing the different types of ANN and selecting the best methods to implement in the different stages of the system developed. Conclusions point to the use of Deep Neural Networks (DNN) with normalized data for the identification of ADL with 85.89% of accuracy, the use of Feedforward neural networks with non-normalized data for the identification of the environments with 86.50% of accuracy, and the use of DNN with normalized data for the identification of standing activities with 100% of accuracy.

This paper focuses on the recognition of Activities of Daily Living (ADL) applying pattern recognition techniques to the data acquired by the accelerometer available in the mobile devices. The recognition of ADL is composed by several stages, including data acquisition, data processing, and artificial intelligence methods. The artificial intelligence methods used are related to pattern recognition, and this study focuses on the use of Artificial Neural Networks (ANN). The data processing includes data cleaning, and the feature extraction techniques to define the inputs for the ANN. Due to the low processing power and memory of the mobile devices, they should be mainly used to acquire the data, applying an ANN previously trained for the identification of the ADL. The main purpose of this paper is to present a new method implemented with ANN for the identification of a defined set of ADL with a reliable accuracy. This paper also presents a comparison of different types of ANN in order to choose the type for the implementation of the final method. Results of this research probes that the best accuracies are achieved with Deep Learning techniques with an accuracy higher than 80%.

Several types of sensors have been available in off-the-shelf mobile devices, including motion, magnetic, vision, acoustic, and location sensors. This paper focuses on the fusion of the data acquired from motion and magnetic sensors, i.e., accelerometer, gyroscope and magnetometer sensors, for the recognition of Activities of Daily Living (ADL) using pattern recognition techniques. The system developed in this study includes data acquisition, data processing, data fusion, and artificial intelligence methods. Artificial Neural Networks (ANN) are included in artificial intelligence methods, which are used in this study for the recognition of ADL. The purpose of this study is the creation of a new method using ANN for the identification of ADL, comparing three types of ANN, in order to achieve results with a reliable accuracy. The best accuracy was obtained with Deep Learning, which, after the application of the L2 regularization and normalization techniques on the sensors data, reports an accuracy of 89.51%., Comment: arXiv admin note: substantial text overlap with arXiv:1711.00124, arXiv:1711.00104; text overlap with arXiv:1711.00096

Most mobile devices include motion, magnetic, acoustic, and location sensors. They allow the implementation of a framework for the recognition of Activities of Daily Living (ADL) and its environments, composed by the acquisition, processing, fusion, and classification of data. This study compares different implementations of artificial neural networks, concluding that the obtained results were 85.89% and 100% for the recognition of standard ADL. Additionally, for the identification of standing activities with Deep Neural Networks (DNN) respectively, and 86.50% for the identification of the environments with Feedforward Neural Networks. Numerical results illustrate that the proposed framework can achieve robust performance from the data fusion of off-the-shelf mobile devices.