IoT applications have been used in many contexts, especially applications on mobile devices. This work presents an attendance checking system by identifying and recognizing human faces on mobile devices using a transfer learning approach. This system includes a mobile application and a web application. These two applications are communicated by using APIs. The mobile application detects human faces by using the camera on that mobile device, then, the face features are extracted using the FaceNet model, and finally, the attendees are identified by computing similarity with existing faces in the database. The proposed system was tested on a dataset with 358 images of 52 employees in our office. Results show that the accuracy is about 93.46% on the test set and 97.06% in the real environment. Thus, this system could be used for checking attendances in several real contexts.

Solving a problem requires knowledge of its context - the information needed to solve it. With the advent of Big Data and ML, huge amounts of data is collected in the desire to develop systems which will then find solutions to specified problems. This requires algorithms to discover and effectively use context hidden within provided data. A system more adept in this task can be regarded as more computationally aware of requirements, data sources and methods that are important to solve given problems. This is a highly desirable property which helps to achieve goals. In this article we present a method for estimating the effectiveness of context usage of machine learning algorithms. It is based on comparison of machine learning models trained on data containing various forms of injected context created with use of noise of varying entropy levels. Finally we give results of using this solution on selected machine learning algorithms and benchmark problems from ICxS Contextual Data repository.

Modeling awareness is an important topic in the computer science as it is closely related to preparing systems that know what is needed (e.g. data accumulated or ignored, effector activated) to achieve a given goal. Preparing tools to build and compare dedicated or general aware computational systems can lead to step-by-step hierarchical construction of intelligent solutions. Within this text we show the relation between awareness, selective attention and contextual systems. Using this as a base we propose basic measures of awareness and present example numerical results obtained for selected contextual neural networks and dedicated, multi-problem benchmark sets. The results allow to quantify awareness in terms of context and selective attention and to propose such solution for use in the general case.

Classification is a data processing technique of a great significance both for native eHealth systems and web telemedicine solutions. In this sense, artificial neural networks have been widely applied in telerehabilitation as powerful tools to process information and acquire a new medical knowledge. But effective analysis of multidimensional heterogeneous medical data, still poses considerable difficulties. It was shown that processing too many data features simultaneously is costly and has some adverse effects on the resulting models classification properties. Therefore, there is a strong need to develop new techniques for selecting features from the very large data sets that include many irrelevant, or redundant features. This work addresses the context-related feature selection problem from medical data by proposing utility of Sigma-if neural network being an effective model of neurology patients's low-level distributed selective attention mechanisms. Our experiments indicate that a context-aware technique can reduce the average cost of medical data acquisition and data processing as well as it can decrease classification error probability resulting in increasing the overall eHealth systems reliability.