Iuliana Marin

Răzvan-Alexandru Smădu, Sebastian-Vasile Echim, Dumitru-Clementin Cercel et al.

Unsupervised Domain Adaptation (UDA) is a popular technique that aims to reduce the domain shift between two data distributions. It was successfully applied in computer vision and natural language processing. In the current work, we explore the effects of various unsupervised domain adaptation techniques between two text classification tasks: fake and hyperpartisan news detection. We investigate the knowledge transfer from fake to hyperpartisan news detection without involving target labels during training. Thus, we evaluate UDA, cluster alignment with a teacher, and cross-domain contrastive learning. Extensive experiments show that these techniques improve performance, while including data augmentation further enhances the results. In addition, we combine clustering and topic modeling algorithms with UDA, resulting in improved performances compared to the initial UDA setup.

Mohammad Rasras, Iuliana Marin, Serban Radu

A smart home can be considered a place of residence that enables the management of appliances and systems to help with day-to-day life by automated technology. In the current paper is described a prototype that simulates a context-aware environment, developed in a designed smart home. The smart home environment has been simulated using three agents and five locations in a house. The context-aware agents behave based on predefined rules designed for daily activities. Our proposal aims to reduce operational cost of running devices. In the future, monitors of health aspects belonging to home residents will sustain their healthy life daily.

Mohammad Rasras, Iuliana Marin, Serban Radu et al.

Human action recognition has become an important research focus in computer vision due to the wide range of applications where it is used. 3D Resnet-based CNN models, particularly MC3, R3D, and R(2+1)D, have different convolutional filters to extract spatiotemporal features. This paper investigates the impact of reducing the captured knowledge from temporal data, while increasing the resolution of the frames. To establish this experiment, we created similar designs to the three originals, but with a dropout layer added before the final classifier. Secondly, we then developed ten new versions for each one of these three designs. The variants include special attention blocks within their architecture, such as convolutional block attention module (CBAM), temporal convolution networks (TCN), in addition to multi-headed and channel attention mechanisms. The purpose behind that is to observe the extent of the influence each of these blocks has on performance for the restricted-temporal models. The results of testing all the models on UCF101 have shown accuracy of 88.98% for the variant with multiheaded attention added to the modified R(2+1)D. This paper concludes the significance of missing temporal features in the performance of the newly created increased resolution models. The variants had different behavior on class-level accuracy, despite the similarity of their enhancements to the overall performance.

Adrian Gheorghiu, Iulian-Marius Tăiatu, Dumitru-Clementin Cercel et al.

The detection and classification of diseases in Robusta coffee leaves are essential to ensure that plants are healthy and the crop yield is kept high. However, this job requires extensive botanical knowledge and much wasted time. Therefore, this task and others similar to it have been extensively researched subjects in image classification. Regarding leaf disease classification, most approaches have used the more popular PlantVillage dataset while completely disregarding other datasets, like the Robusta Coffee Leaf (RoCoLe) dataset. As the RoCoLe dataset is imbalanced and does not have many samples, fine-tuning of pre-trained models and multiple augmentation techniques need to be used. The current paper uses the RoCoLe dataset and approaches based on deep learning for classifying coffee leaf diseases from images, incorporating the pix2pix model for segmentation and cycle-generative adversarial network (CycleGAN) for augmentation. Our study demonstrates the effectiveness of Transformer-based models, online augmentations, and CycleGAN augmentation in improving leaf disease classification. While synthetic data has limitations, it complements real data, enhancing model performance. These findings contribute to developing robust techniques for plant disease detection and classification.

Myssar Jabbar Hammood Al-Battbootti, Iuliana Marin, Nicolae Goga et al.

The vast network of oil and gas transmission pipelines requires periodic monitoring for maintenance and hazard inspection to avoid equipment failure and potential accidents. The severe COVID-19 pandemic situation forced the companies to shrink the size of their teams. One risk which is faced on-site is represented by the uncontrolled release of flammable oil and gas. Among many inspection methods, the unmanned aerial vehicle system contains flexibility and stability. Unmanned aerial vehicles can transfer data in real-time, while they are doing their monitoring tasks. The current article focuses on unmanned aerial vehicles equipped with optical sensing and artificial intelligence, especially image recognition with deep learning techniques for pipeline surveillance. Unmanned aerial vehicles can be used for regular patrolling duties to identify and capture images and videos of the area of interest. Places that are hard to reach will be accessed faster, cheaper and with less risk. The current paper is based on the idea of capturing video and images of drone-based inspections, which can discover several potential hazardous problems before they become dangerous. Damage can emerge as a weakening of the cladding on the external pipe insulation. There can also be the case when the thickness of piping through external corrosion can occur. The paper describes a survey completed by experts from the oil and gas industry done for finding the functional and non-functional requirements of the proposed system.

Iuliana Marin, Ioana-Andreea Dinescu, Teodora-Coralia Deleanu et al.

Deficit of attention, anxiety, sleep disorders are some of the problems which affect many persons. As these issues can evolve into severe conditions, more factors should be taken into consideration. The paper proposes a conception which aims to help students to enhance their brain performance. An electrocephalogram headset is used to trigger the brainwaves, along with a web application which manages the input data which comes from the headset and from the user. Factors like current activity, mood, focus, stress, relaxation, engagement, excitement and interest are provided in numerical format through the use of the headset. The users offer information about their activities related to relaxation, listening to music, watching a movie, and studying. Based on the analysis, it was found that the users consider the application easy to use. As the users are more equilibrated emotionally, their results are improved. This allowed the persons to be more confident on themselves. In the case of students, the neurofeedback can be studied for the better sport and artistic performances, including the case of the attention deficit hyperactivity disorder. Aptitudes for a subject can be determined based on the relevant generated brainwaves. The learning environment is an important factor during the analysis of the results. Teachers, professors, students and parents can collaborate and, based on the gathered data, new teaching methods can be adopted in the classroom and at home. The proposed solution can guide the students while studying, as well as the persons who wish to be more productive while solving their tasks.

Iuliana Marin, Myssar Jabbar Hammood Al-Battbootti, Nicolae Goga

When it is tried to control drones, there are many different ways through various devices, using either motions like facial motion, special gloves with sensors, red, green, blue cameras on the laptop or even using smartwatches by performing gestures that are picked up by motion sensors. The paper proposes a work on how drones could be controlled using brainwaves without any of those devices. The drone control system of the current research was developed using electroencephalogram signals took by an Emotiv Insight headset. The electroencephalogram signals are collected from the users brain. The processed signal is then sent to the computer via Bluetooth. The headset employs Bluetooth Low Energy for wireless transmission. The brain of the user is trained in order to use the generated electroencephalogram data. The final signal is transmitted to Raspberry Pi zero via the MQTT messaging protocol. The Raspberry Pi controls the movement of the drone through the incoming signal from the headset. After years, brain control can replace many normal input sources like keyboards, touch screens or other traditional ways, so it enhances interactive experiences and provides new ways for disabled people to engage with their surroundings.

Hasan Naji, Iuliana Marin, Nicolae Goga et al.

The purpose of this project is to build a system that will quickly track the location of a stolen vehicle, thereby reducing the cost and effort of police. Moreover, the vehicle's computer system can be controlled remotely by the owners of the vehicle or police. More precisely, the goal of this work is to design a, develop remote control of the vehicle, and find the locations with Latitude (LAT) and Longitude (LONG).

Andrei Vasilateanu, Nicolae Goga, Elena-Alice Tanase et al.

Enterprise knowledge is a key asset in the competing and fast-changing corporate landscape. The ability to learn, store and distribute implicit and explicit knowledge can be the difference between success and failure. While enterprise knowledge management is a well-defined research domain, current implementations lack orientation towards small and medium enterprise. We propose a semantic search engine for relevant documents in an enterprise, based on automatic generated domain ontologies. In this paper we focus on the component for ontology learning and population.

Iuliana Marin

The COVID pandemic and the measures which were taken had effect over the mental health of persons. The current paper proposes a concept that supports the performance of students by analyzing three ways of distance learning, namely text, text and illustrations, including charts, video. An electroencephalogram headset allows the detection of brainwaves and the developed web application enhances the process of distance learning. The electrodes of the headset are placed at contact with the user head and monitor the activity of the left and right frontal regions, along with the temporal lobe. Mood, focus, stress, relaxation, engagement, excitement and interest are triggered as numerical values by using the headset. The users provide information about their daily activities, including learning and evaluation processes. According to the study, users had the highest long term attention while using text and illustrations, followed by watching videos. This is caused by the fact that the text contained the code for the programs which were presented in the video. Also, the users feel comfortable while using the application and they started to pay more attention to the connection between stress, health, education and well being. The results triggered by the headset had higher values while students studied for the first time using videos. When they wanted to remember the information, the text and illustrations way of learning was the best option. Based on the study outcomes, the instructional design can be enhanced. Moreover, the results improved as the students became more equilibrated and confident in themselves. Teachers, professors and parents are able to collaborate and enhance training. While studying online under lockdown, students have found the proposed solution to be good because their inner state influences their productivity while solving problems.