Oslo: Møte 21. februar 2018
onsdag, 21 feb 2018
18:00 – 22:00
Big data for norsk verdiskapning
Big Data in the Media Industry
Jon Atle Gulla, professor, NTNU
The media industry is undergoing a deep transformation from paper-based newspapers to online news experiences. Readers now have access to a plethora of news stories from online media houses and social sites from around the world in real-time. However, news stories have traditionally been hard to analyze due to their fragmented and subjective nature, their ambiguities and inconsistences, and the complexities of natural language text in general. Technologies from Big Data and Artificial Intelligence help us extract meaningful content from news stories and deal with massive streams of real-time news. In this presentation we will see how news content can be personalized and localized in real-time to give the reader a feeling of what is going on right now in her neighborhood. We show how trends and sentiments can be extracted from millions of news posts, and also discuss how new services like targeted advertising affect the entire business model of the media industry.
Industrial Big Data
Harald Martens, forskningssjef, Idletechs AS og professor II, NTNU
With powerful new measuring devices and data storage tools now available, radically new opportunities open up in industry. However, before the torrents of informative data from low-cost, high-speed monitoring sensors like thermal cameras, spectrophotometers and accelerometers can be put to real use, better data modelling methods are needed.
The average car driver does not have to know how the brakes work. But the car-producer and the mechanical workshop can tell how the brakes work. Likewise, the average user of an AI application does not have to know how the AI application works. But today NOBODY can tell how the AI application works, if it is based on a “black box” machine learning method. Industry cannot rely on black box operation. How, then, to handle Industrial Big Data?
The lecture will focus on some tools that simplify and enhance the use of modern sensors, SCADA data and Data Warehouse structures in industry:
- Traditional selectivity in raw measurements is no longer needed. Instead, the technique of multivariate calibration is a generic tool from the field of chemometrics to convert high-speed but non-selective measurements into precise, accurate and understandable information.
- Continuous machine learning without the black box: Torrents of Quantitative Big Data – i.e. “ever-lasting” streams of high-dimensional measurements can be overwhelming. But they can easily be converted into compact, quantitative and interpretable information, with fewer, but more sensitive error warnings, using new, generic industrially data modelling tools from the field of “Big Data Cybernetics”.
Big Data meets Graph Signal Processing
Baltasar Beferull-Lozano, professor, Universitetet i Agder
There has been significant recent progress in the development of signal processing tools that operate directly over graph-structured data. Many datasets in multiple real applications can be modeled flexibly in terms of the so-called graph signals, where each node of the graph represents essentially one or several data time-series and the links connecting the nodes represent some type of space-time dependencies among the various data time-series. The emergence of very diverse and ubiquitous computing, communicating and sensing devices has led to an era where huge amounts of data (big data) are generated constantly. Examples of applications include industrial plants, cyber-physical networked sensor systems, water distribution networks, connectivity brain networks, finance networked systems and social networks, among others. This talk will provide an overview about the recent area of Graph Signal Processing, exploring recent results, challenges and applications. We will illustrate the key concepts and cover different data processing methods for machine learning tasks over graph signals, their computational efficiency, and their impact in several important applications.
Møtet er åpent for alle interesserte.
De som vil delta på middagen etter foredraget må melde seg på til firstname.lastname@example.org innen mandag 19. februar. Egenandelen for middagen er kr 400, og påmeldte vil få tilsendt faktura.
Det Norske Videnskaps-Akademi