A5009239738- Advanced Image and Video Retrieval Techniques
- Text and Document Classification Technologies
- Image Retrieval and Classification Techniques
- Face and Expression Recognition
- Air Quality Monitoring and Forecasting
- Multimodal Machine Learning Applications
- Machine Learning and Data Classification
- Topic Modeling
- Video Analysis and Summarization
- Natural Language Processing Techniques
- Handwritten Text Recognition Techniques
- Data Mining Algorithms and Applications
- Privacy, Security, and Data Protection
- Machine Learning and ELM
- Impact of Light on Environment and Health
- Ethics and Social Impacts of AI
- Domain Adaptation and Few-Shot Learning
- Text Readability and Simplification
- Imbalanced Data Classification Techniques
- Privacy-Preserving Technologies in Data
- Robotics and Sensor-Based Localization
- Misinformation and Its Impacts
- Space Exploration and Technology
- Speech and dialogue systems
- Spam and Phishing Detection
Aristotle University of Thessaloniki
2011-2020
Information Technologies Institute
2012-2018
Hella (Germany)
2018
Centre for Research and Technology Hellas
2014-2018
This paper deals with content-based large-scale image retrieval using the state-of-the-art framework of VLAD and Product Quantization proposed by Jegou as a starting point. Demonstrating an excellent accuracy-efficiency trade-off, this has attracted increased attention from community numerous extensions have been proposed. In work, we make in-depth analysis that aims at increasing our understanding its different processing steps boosting overall performance. Our involves evaluation (both...
Machine learning bias and fairness have recently emerged as key issues due to the pervasive deployment of data-driven decision making in a variety sectors services. It has often been argued that unfair classifications can be attributed training data, but previous attempts 'repair' data led limited success. To circumvent shortcomings prevalent repairing approaches, such those weight samples sensitive group (e.g. gender, race, financial status) based on their misclassification error, we...
Streams of objects that are associated with one or more labels at the same time appear in many applications. However, stream classification multi-label data is largely unexplored. Existing approaches try to tackle problem by transferring traditional single-label practices domain. Nevertheless, they fail consider some unique properties such as within and between class imbalance multiple concept drift. To deal these challenges, this paper proposes a novel multilabel approach employs two...
Information sharing in online social networks is a daily practice for billions of users. The process facilitates the maintenance users' ties but also entails privacy disclosure relation to other users and third parties. Depending on intentions latter, this can become risk. It thus important propose tools that empower their relations parties connected them. As part USEMP, coordinated research effort aimed at user empowerment, we introduce system performs privacy-aware classification images....
Although air pollution is one of the most significant environmental factors posing a threat to human health worldwide, quality data are scarce or not easily accessible in European countries. The current work aims develop centralized hub that enables citizens contribute monitoring. In this work, from official monitoring stations combined with estimates sky-depicting photos and low-cost sensing devices build on their own so receive improved information about they breathe. Additionally, fusion...
This paper proposes a machine learning approach to part-of-speech tagging and named entity recognition for Greek, focusing on the extraction of morphological features classification tokens into small set classes entities. The architecture model that was used is introduced. greek version spaCy platform added source code, feature did not exist before our contribution, building models. Additionally, part speech tagger trained can detect morphology performs higher than state-of-the-art results...
Results returned by commercial image search engines should include relevant and diversified depictions of queries in order to ensure good coverage users' information needs. While relevance has drastically improved recent years, diversity is still an open problem. In this paper we propose a reranking method that could be implemented on top such provide better balance between diversity. Our formulates the problem as optimization utility function jointly considers main contribution replacement...
The study of efficient image representations has attracted significant interest due to the computational needs large-scale applications. In this paper we performance recently proposed VLAD method for aggregating local descriptors when combined with SURF features, in domain search. experiments show that features are used as descriptors, attains better compared using SIFT features. We also how average number extracted per affects and by controlling able adjust trade off between feature...
A key challenge in information theoretic feature selection is to estimate mutual expressions that capture three desirable terms—the relevancy of a with the output, redundancy and complementarity between groups features. The becomes more pronounced multi-target problems, where output space multi-dimensional. Our work presents an algorithm captures these terms suitable for well-known prediction settings multi-label/dimensional classification multivariate regression. We achieve this by...
The public availability of large-scale multimedia collections, such as YFCC, facilitates the evaluation image retrieval approaches in real-life conditions. However, due to their size, creation exhaustive ground truth would require huge annotation effort, even for limited sets queries. This paper investigates whether it is possible estimate performance absence manually created data. Our hypothesis that leverage existing weak user annotations (tags) automatically build To test our hypothesis,...
Multi-target regression is concerned with the prediction of multiple continuous target variables using a shared set predictors. Two key challenges in multi-target are: (a) modelling dependencies and (b) scalability to large output spaces. In this paper, new method proposed that tries jointly address these via novel problem transformation approach. The method, called MRQ, based on idea quantizing space order transform targets into one or more discrete ones. Learning transformed naturally...