A5030362528- Context-Aware Activity Recognition Systems
- Music and Audio Processing
- Phonocardiography and Auscultation Techniques
- COVID-19 diagnosis using AI
- Non-Invasive Vital Sign Monitoring
- Ethics and Social Impacts of AI
- Machine Learning in Healthcare
- Time Series Analysis and Forecasting
- Physical Activity and Health
- Topic Modeling
- Speech Recognition and Synthesis
- Speech and Audio Processing
- Human Pose and Action Recognition
- Privacy, Security, and Data Protection
- Obstructive Sleep Apnea Research
- Advanced Text Analysis Techniques
- IoT and Edge/Fog Computing
- Natural Language Processing Techniques
- Mental Health Research Topics
- Machine Learning and Data Classification
- Chronic Obstructive Pulmonary Disease (COPD) Research
- Technology Use by Older Adults
- Respiratory and Cough-Related Research
- Psychological Well-being and Life Satisfaction
- Speech and dialogue systems
Nokia (United Kingdom)
2022-2025
University of Cambridge
2018-2024
Aristotle University of Thessaloniki
2017-2019
Ionian University
2014-2017
Telefonica Research and Development
2017
University College London
2017
Audio signals generated by the human body (e.g., sighs, breathing, heart, digestion, vibration sounds) have routinely been used clinicians as indicators to diagnose disease or assess progression. Until recently, such were usually collected through manual auscultation at scheduled visits. Research has now started use digital technology gather bodily sounds from stethoscopes) for cardiovascular respiratory examination, which could then be automatic analysis. Some initial work shows promise in...
Abstract To identify Coronavirus disease (COVID-19) cases efficiently, affordably, and at scale, recent work has shown how audio (including cough, breathing voice) based approaches can be used for testing. However, there is a lack of exploration biases methodological decisions impact these tools’ performance in practice. In this paper, we explore the realistic audio-based digital testing COVID-19. investigate this, collected large crowdsourced respiratory dataset through mobile app,...
This study examines the clinical decision support systems in healthcare, particular about prevention, diagnosis and treatment of respiratory diseases, such as Asthma chronic obstructive pulmonary disease. The empirical pulmonology a representative sample (n = 132) attempts to identify major factors that contribute these diseases. Machine learning results show disease’s case, Random Forest classifier outperforms other techniques with 97.7 per cent precision, while most prominent attributes...
Machine learning and deep have shown great promise in mobile sensing applications, including Human Activity Recognition. However, the performance of such models real-world settings largely depends on availability large datasets that captures diverse behaviors. Recently, studies computer vision natural language processing leveraging massive amounts unlabeled data enables par with state-of-the-art supervised models. In this work, we present SelfHAR, a semi-supervised model effectively learns...
The development of fast and accurate screening tools, which could facilitate testing prevent more costly clinical tests, is key to the current pandemic COVID-19. In this context, some initial work shows promise in detecting diagnostic signals COVID-19 from audio sounds. paper, we propose a voice-based framework automatically detect individuals who have tested positive for We evaluate performance proposed on subset data crowdsourced our app, containing 828 samples 343 participants. By...
The INTERSPEECH 2021 Computational Paralinguistics Challenge addresses four different problems for the first time in a research competition under well-defined conditions: In COVID-19 Cough and Speech Sub-Challenges, binary classification on infection has to be made based coughing sounds speech; Escalation Sub-Challenge, three-way assessment of level escalation dialogue is featured; Primates species vs background need classified.We describe baseline feature extraction, classifiers 'usual'...
Are there universal patterns in musical preferences? To address this question, we built on theory and research personality, cultural, music psychology to map the terrain of preferences for Western using data from 356,649 people across six continents. In Study 1 (N = 284,935), participants 53 countries completed a genre favorability measure, 2 71,714), 36 an audio-based measure preferential reactions music. Both studies included self-report measures Big Five personality traits demographics....
Human Activity Recognition (HAR) constitutes one of the most important tasks for wearable and mobile sensing given its implications in human well-being health monitoring. Motivated by limitations labeled datasets HAR, particularly when employed healthcare-related applications, this work explores adoption adaptation SimCLR, a contrastive learning technique visual representations, to HAR. The use objectives causes representations corresponding views be more similar, those non-corresponding...
Limited availability of labeled data for machine learning on multimodal time-series extensively hampers progress in the field. Self-supervised (SSL) is a promising approach to learn representations without relying labels. However, existing SSL methods require expensive computations negative pairs and are typically designed single modalities, which limits their versatility. We introduce CroSSL (Cross-modal SSL), puts forward two novel concepts: masking intermediate embeddings produced by...
Wearable technologies enable continuous monitoring of various health metrics, such as physical activity, heart rate, sleep, and stress levels. A key challenge with wearable data is obtaining quality labels. Unlike modalities like video where the videos themselves can be effectively used to label objects or events, do not contain obvious cues about manifestation users usually require rich metadata. As a result, noise become an increasingly thorny issue when labeling data. In this paper, we...
Large language models (LLMs) show promise for health applications when combined with behavioral sensing data. Traditional approaches convert sensor data into text prompts, but this process is prone to errors, computationally expensive, and requires domain expertise. These challenges are particularly acute processing extended time series While foundation (TFMs) have recently emerged as powerful tools learning representations from temporal data, bridging TFMs LLMs remains challenging. Here, we...
The adoption of multisensor wearables presents the opportunity longitudinal monitoring sleep in large populations. Personalized yet device-agnostic algorithms can sidestep laborious human annotations and objectify cross-cohort comparisons. We developed tested a heart rate-based algorithm that captures inter- intra-individual differences free-living conditions does not require input. evaluated it on four study cohorts using different research- consumer-grade devices for over 2000 nights....
Recent work has shown the potential of using audio data (eg, cough, breathing, and voice) in screening for COVID-19. However, these approaches only focus on one-off detection detect infection, given current sample, but do not monitor disease progression Limited exploration been put forward to continuously COVID-19 progression, especially recovery, through longitudinal data. Tracking characteristics patterns recovery could bring insights lead more timely treatment or adjustment, as well...
The COVID-19 pandemic has caused massive humanitarian and economic damage. Teams of scientists from a broad range disciplines have searched for methods to help governments communities combat the disease. One avenue machine learning field which been explored is prospect digital mass test can detect infected individuals’ respiratory sounds. We present summary results INTERSPEECH 2021 Computational Paralinguistics Challenges: Cough, (CCS) Speech, (CSS).
Common approaches to text categorization essentially rely either on n-gram counts or word embeddings. This presents important difficulties in highly dynamic quickly-interacting environments, where the appearance of new words and/or varied misspellings is norm. A paradigmatic example this situation abusive online behavior, with social networks and media platforms struggling effectively combat uncommon non-blacklisted hate words. To better deal these issues those fast-paced we propose using...
Experience sampling has long been the established method to sample people's mood in order assess their mental state. Smartphones start be used as experience tools for health state they accompany individuals during day and can therefore gather in-the-moment data. However, granularity of data needs traded off with level interruption these introduce. As a consequence collected this technique is often sparse. This obviated by use passive sensing addition reports, however, adds additional noise.
Self-supervised learning (SSL) has shown remarkable performance in computer vision tasks when trained offline. However, a Continual Learning (CL) scenario where new data is introduced progressively, models still suffer from catastrophic forgetting. Retraining model scratch to adapt newly generated time-consuming and inefficient. Previous approaches suggested re-purposing self-supervised objectives with knowledge distillation mitigate forgetting across tasks, assuming that labels all are...
Large language models (LLMs) have demonstrated remarkable generalization and across diverse tasks, leading individuals to increasingly use them as personal assistants due their emerging reasoning capabilities. Nevertheless, a notable obstacle emerges when including numerical/temporal data into these prompts, such sourced from wearables or electronic health records. LLMs employ tokenizers in input that break down text smaller units. However, are not designed represent numerical values might...
Smartphones have started to be used as self reporting tools for mental health state they accompany individuals during their days and can therefore gather temporally fine grained data. However, the analysis of reported mood data offers challenges related non-homogeneity assessment among due complexity feeling scales, well noise sparseness reports when collected in wild. In this paper, we propose a new end-to-end ML model inspired by video frame prediction machine translation, that forecasts...