Multimedia Computing Group
The Multimedia Computing (MMC) Group develops algorithms for enriching, accessing, and searching large quantities of data. Such algorithms lie at the core of tomorrows’ search engines and large-scale recommender systems. The group sets its focus on developing systems that are oriented to the needs of users, and that solve the challenges faced by large-scale online content and service providers. Multimedia data analytics also has applications in the full range of fields that benefit from data science, including health, telecom, and geosciences.
The group has a track record of developing technologies that make possible optimized interaction with large collections of multimedia data (e.g., images, video, and music) in real-world contexts (e.g., within social networks). Our work requires a combination of mathematical models, machine learning techniques, and practical skills in algorithm development and evaluation. The members of the MMC group share expertise in multimedia information retrieval, recommender systems, multimedia signal processing, social network analysis, human computation (crowdsourcing) and quality of experience. Collaborations include joint work with researchers from Yahoo Labs, Telefónica Research, Microsoft Research and Google.
The work being carried out in the MMC Group encompasses a broad palette of research directions including:
- Multimedia content analysis and search
- Semantics extraction from multimedia data
- Multi-modal query expansion
- Multi-source search result reranking
- Multimedia information retrieval in a social network context
- Modeling information propagation and relationships in social networks
- Collaborative recommender systems
- Social recommendation
- Interaction with multimedia content
- (Affective) User profiling
- User (search/uploader) intent
- Query failure prediction
- Quality of multimedia experience
- Multimedia content management
- Multimedia databases and dataspaces
- Entity retrieval
- Human speech processing inspired automatic speech recognition:
- Building speech technology for under-resourced languages and languages without a common written language.
- Using knowledge about human speech processing to build help improve ASR algorithms.
- Building computational models of human speech processing using techniques from ASR.
10 augustus 2020
Graph Signal Processing: Connections to Distributed Optimization and Deep Learning
Graph Signal Processing: Connections to Distributed Optimization and Deep Learning. A tutorial by Elvin Isufi at the International Conference on Signal Processing and Communications July 19 - 23, 2020
01 maart 2020
Graph-Time Spectral Analysis for Atrial Fibrillation
Atrial fibrillation is a clinical arrhythmia with multifactorial mechanisms still unresolved. Time- frequency analysis of epicardial electrograms has been investigated to study atrial fibrillation. How- ever, deeper understanding of atrial fibrillation can be achieved if the spatial dimension can be in- corporated. Unfortunately, the physical models describing the spatial relations of atrial fibrillation signals are complex and non-linear; hence, the conventional signal processing techniques to study electrograms in the joint space, time, and frequency domain are less suitable. In this study, we wish to put forward a radically different approach to analyze atrial fibrillation with a higher-level model.
16 februari 2020
State-Space Network Topology Identification from Partial Observations
Identifying how entities are connected by observing the evolution of time series is an important problem is biology, finance, and sensor networks. The problem known as "network topology identification (NTI)" is both challenging to solve and difficult to be characterised theoretically. In our work, we undust some old tools from system identification literature and show how they can be used for NTI. This link allowed us to provide theoretical guarantees on when a topology is identifiable from partial measurements and to develop an algorithm for retreiving it. When the theoretical conditions are met, our proposed algorithm finds the #true network being responsible for the dynamics.
26 november 2019
Odette Scharenborg keynote speaker at Speech Technology Deep Dive
On Wednesday November 27, 2019, Odette Scharenborg is a keynote speaker and discussant at the Speech Technology Deep Dive of the Dutch ministries of Justice and Internal Affairs. During TechNovember, which lasts the entire month of November, 12 so-called "deepdives" take place. In such a deep dive, a scientist by name and fame always plunges into depth with the participants from both departments.
26 november 2019
Door mijn Siri-schaamte heb ik voorlopig geen butler in mijn broekzak
A digital assistant who arranges things for you if you ask. It sounds like an ideal world in which you always carry a help, but in reality you constantly end up in a Babylonian confusion of words. Odette Scharenborg was interviewed for a background article on speech technology for the Dutch news website NU.nl.
10 november 2019
Article in IEEE Transactions on Signal Processing by Elvin Isufi
Article in IEEE Transactions on Signal Processing: Controllability of Bandlimited Graph Processes Over Random Time-Varying Graphs by E. Isufi