A5020175228- Context-Aware Activity Recognition Systems
- Anomaly Detection Techniques and Applications
- Human Mobility and Location-Based Analysis
- Privacy-Preserving Technologies in Data
- Indoor and Outdoor Localization Technologies
- Time Series Analysis and Forecasting
- Gait Recognition and Analysis
- Healthcare Technology and Patient Monitoring
- Industrial Vision Systems and Defect Detection
- Human Pose and Action Recognition
- Mobile Crowdsensing and Crowdsourcing
- Video Surveillance and Tracking Methods
- Fault Detection and Control Systems
- Machine Learning in Healthcare
- Domain Adaptation and Few-Shot Learning
- Advanced Statistical Process Monitoring
- Speech Recognition and Synthesis
- Data-Driven Disease Surveillance
- Internet Traffic Analysis and Secure E-voting
Saints Cyril and Methodius University of Skopje
2019-2023
Hasso Plattner Institute
2023
University of Potsdam
2023
Jožef Stefan International Postgraduate School
2020-2022
Jožef Stefan Institute
2020-2022
Convolution Neural Network (CNN) filters learned on one domain can be used as feature extractors another similar domain. Transferring allow reusing datasets across domains and reducing labelling costs. In this paper, four activity recognition were analyzed to study the effects of transferring datasets. A spectro-temporal ResNet was implemented a deep, end-to-end learning architecture. We number transferred CNN residual blocks with respect size target-adaptation data. The analysis showed that...
The past decade has seen substantial growth in the prevalence and capabilities of wearable devices. For instance, recent human activity recognition (HAR) research explored using devices applications such as remote monitoring patients, detection gait abnormalities, cognitive disease identification. However, data collection poses a major challenge developing HAR systems, especially because need to store at central location. This raises privacy concerns makes continuous difficult expensive due...
The SHL recognition challenge 2020 was an open competition in activity where the participants were tasked with recognizing eight different modes of locomotion and transportation smartphone sensors. main challenges that training data recorded by a person than validation test data, location unknown to participants. We, team "Third time's charm", tackled first attempting identify persons clustering, then performed cluster/person-specific feature selection build separate classifier for each...
From 2018 to 2021, the Sussex-Huawei Locomotion-Transportation Recognition Challenge presented different scenarios in which participants were tasked with recognizing eight modes of locomotion and transportation using sensor data from smartphones. In 2019, main challenge was one location recognize activities sensors another location, while following year, person other persons. We use these two as a framework analyze effectiveness components machine-learning pipeline for activity recognition....
Human Activity Recognition (HAR) from wearable sensors has gained significant attention in the last few decades, largely because of potential healthcare benefits. For many years, HAR was done using classical machine learning approaches that require extraction features. With resurgence deep learning, a major shift happened and at moment, researchers are mainly investigating different kinds neural networks. However, comes with challenge having access to large amounts labeled examples, which...
Accurate and comprehensive nursing documentation is essential to ensure quality patient care. To streamline this process, we present SONAR, a publicly available dataset of activities recorded using inertial sensors in home. The includes 14 sensor streams, such as acceleration angular velocity, 23 by caregivers five for 61.7 hours. wore the they performed their daily tasks, allowing continuous monitoring activities. We additionally provide machine learning models that recognize given data. In...
In light of events such as the recent pandemic and many potential applications in fields social sciences, healthcare, architecture, detection interactions or proximity between people has become increasingly important. this context, paper investigates limitations a machine learning-based approach that detects two devices based on WiFi BLE fingerprints their radio environments. More specifically, (i) we compare use rudimentary set features an extended, more complex features, (ii) investigate...
Speech-based machine learning models that can distinguish between a healthy cognitive state and different stages of decline would enable more appropriate timely treatment patients. However, their development is often hampered by data scarcity. Federated Learning (FL) potential solution could entities with limited voice recordings to collectively build effective models. Motivated this, we compare centralised, local, federated for building speech-based discern Alzheimer's Disease, Mild...
Today's fast paced industrial production requires automation at multiple steps during its process. Involving humans the quality control inspection provides high degree of confidence that end products are with best quality. Workers involved in process may have an impact on capacity by lowering throughput, depending complexity time is carried out, which a time-critical operation, or after completed. Companies striving to fully automate their stages and it comes naturally focus using various...
Wearable devices have the ability to generate vast amounts of data that can be put use in a multitude applications, particularly field e-health. However, potential invasion privacy comes with utilizing personal collected by these cannot overlooked. Federated Learning (FL) is promising solution this issue allows models trained decentralized manner while keeping user on their own devices. This approach effectively minimizes risk breaches and has employed variety applications where protection...
Wireless devices such as smartphones, wireless bracelets or smartwatches are now often used by a single person. They can therefore be to model human contact, proximity each other, social networking. In light of events the recent pandemic interactions between people have become increasingly important. this context, paper explores limitations machine learning-based approach that detects two (up about metres) based on WiFi and BLE (Bluetooth Low Energy) fingerprints their radio environments....