642 research outputs found
Cross-Media Knowledge Extraction in the Car Manufacturing Industry
No full textIn this paper, we present a novel framework for machine
learning-based cross-media knowledge extraction.
The framework is specifically designed to handle documents
composed of three types of media – text, images and raw
data – and to exploit the evidence for an extracted fact
from across the media. We validate the framework by applying
it in the design and development of cross-media extraction
systems in the context of two real-world use cases in the car manufacturing industry. Moreover, we show that in these use cases the cross-media approach effectively improves system extraction accuracy
Evidence-driven policy-making using heterogeneous data - The case of a controlled parking system in Thessaloniki
No full textPolicy making in local public administrations is still largely based on intuition rather than being backed up by data and evidence. The goal of this work is to introduce the methodology and software tools for contributing towards transforming the existing intuition-based paradigm of policy making into an evidence-driven approach enabled by heterogeneous sources of data already available in the city. More specifically, methods for data collection, efficient data storage and data analysis are implemented to measure the economic activity, assess the environmental impact and evaluate the social consequences of certain policy decisions. Subsequently, the extracted pieces of evidence are used to inform, advise, monitor, evaluate and revise the decisions made by policy planners. Our contribution in this work is on outlining and deploying an easily extendable system architecture to harmonize and analyze heterogeneous data sources in ways that are found to be useful for policy makers. For evaluating this architecture, we examine the case of a controlled parking system (CPS) in the city of Thessaloniki and try to optimize its operation by balancing effectively between economic growth, environmental protection and citizen satisfaction.Anastasios Papazoglou Chalikias, Ioannis Tsampoulatidis, Filareti Tsalakanidou, Spiros Nikolopoulos, Ioannis Kompatsiaris, Nicos Komninos, Konstantinos Doudouliakis, Georgios Papastergios, Petros Papafilis, Sophia Karkaletsi, and Charalambos Chatzis, "Evidence-driven policy-making using heterogeneous data - The case of a controlled parking system in Thessaloniki", Data & Policy Journal, Vol. 2, E 15, November 2020. (DOI: https://doi.org/10.1017/dap.2020.15
The representation of speech in deep neural networks
No full textIn this paper, we investigate the connection between how people understand speech and how speech is understood by a deep neural network. A naïve, general feed-forward deep neural network was trained for the task of vowel/consonant classification. Subsequently, the representations of the speech signal in the different hidden layers of the DNN were visualized. The visualizations allow us to study the distance between the representations of different types of input frames and observe the clustering structures formed by these representations. In the different visualizations, the input frames were labeled with different linguistic categories: sounds in the same phoneme class, sounds with the same manner of articulation, and sounds with the same place of articulation. We investigate whether the DNN clusters speech representations in a way that corresponds to these linguistic categories and observe evidence that the DNN does indeed appear to learn structures that humans use to understand speech without being explicitly trained to do so.Accepted author manuscriptMultimedia Computin
The step project:societal and political engagement of young people in environmental issues
Get PDFDecisions on environmental topics taken today are going to have long-term consequences that will affect future generations. Young people will have to live with the consequences of these decisions and undertake special responsibilities. Moreover, as tomorrow’s decision makers, they themselves should learn how to negotiate and debate issues before final decisions are made. Therefore, any participation they can have in environmental decision making processes will prove essential in developing a sustainable future for the community.However, recent data indicate that the young distance themselves from community affairs, mainly because the procedures involved are ‘wooden’, politicians’ discourse alienates the young and the whole experience is too formalized to them. Authorities are aware of this fact and try to establish communication channels to ensure transparency and use a language that speaks to new generations of citizens. This is where STEP project comes in.STEP (www.step4youth.eu) is a digital Platform (web/mobile) enabling youth Societal and Political e-Participation in decision-making procedures concerning environmental issues. STEP is enhanced with web/social media mining, gamification, machine translation, and visualisation features.Six pilots in real contexts are being organised for the deployment of the STEP solution in 4 European Countries: Italy, Spain, Greece, and Turkey. Pilots are implemented with the direct participation of one regional authority, four municipalities, and one association of municipalities, and include decision-making procedures on significant environmental questions.</p
HierArtEx: Hierarchical Representations and Art Experts Supporting the Retrieval of Museums in the Metaverse
No full textThe improvements in Virtual Reality technologies are bringing more attention to the Metaverse and the nearly unlimited experiences available there. Among these, digital museums have seen an increase in the number of yearly visitors, especially after the COVID-19 pandemics. However, no tools are available to support the user in the searching process. To this end, we start investigating the Text-to-Museum retrieval task, involving museums composed of many rooms enriched by multimedia elements affecting their relevance to the user query. To model this complex type of data, we design HierArtEx, which leverages hierarchical representations to model the whole museum, while combining generic and art-specific knowledge for capturing the visual contents of each single room. We validate its effectiveness on Museums3k, a large dataset that we collect, containing 3000 realistic museums each annotated by a description of its contents. Moreover, qualitative analyses confirm favorable results and their alignment with real user queries, while also highlighting the shortcomings of standard evaluation protocols in retrieval, as they fail to capture all relevant museums
- …
