Big Thrones: The power of ML // Video Reco: Find what you were (not) looking for

For this edition focused on TV content, we're very happy to be hosted at ProSiebenSat.1. They will sponsor the venue, the food and the drinks!

Getting there is easy: Just take the SBahn S8 to station Unterföhring and you're there!

As there will be a security check at the entrance I'd like to kindly ask everyone to Register on the Eventbrite page (https://www.eventbrite.de/e/machine-learning-meetup-tickets-27485357444) so I have a master list of names. IF YOUR NAME IS NOT ON THE LIST YOU WON'T GET IN. So make sure to register, pretty please.

We are lucky to have 2 very exciting talks related to multimedia entertainment.

Dr. Goldberg and Dr. Yachdav will talk about how they are using Machine Learning to predict likelyhood of death in Game of Thrones.

Dr Anca Tudoran will give a talk on how to do recommendation for a Video Hyperlinking system, in order to improve experience of viewers.

-----------------------------------------------------

AGENDA:

18:30: Doors open

18:30 - 19:00: Networking / Drink / Food

19:00 - 20:00: Big Thrones: The power of Machine Learning

20:00 - 20:15: Break

20:15 - 21:00: Video Recommendations: Find what you were (not) looking for

21:00 - 22:00: Networking

22:00 : Doors close

-----------------------------------------------------

Title: Big Thrones: The power of Machine Learning

Author: Dr. Guy Yachdav, Dr. Tatyana Goldberg

Abstract: Early this year a group of students (and their mentors) from the TU Munich decided to take on a unique challenge - use machine learning to predict the likelihood of death for the characters in the HBO hit show “Game of Thrones”. In this talk we will give an insight into a 50 days journey that ended up with a result that has been featured in news media across the world, including Time, The Guardian, BBC and more. If you happen to be a data geek or a Game ofThrones nerd (or both) this talk is for you!

Bios:

Dr. Guy Yachdav believes that data is the key to understanding and solving problems that affect every single aspect of our lives. As a researcher, Guy worked with genomic data, developing software that helps scientists understand the way proteins carry out their function. He also created data analysis platforms that analyze the role of bacteria colonies in the digestive system, led an effort to create an open source biological data visualization tool, and built an online computational workflow that is used by tens of thousands of scientists. As a technological entrepreneur, Guy founded several startup companies including Biosof LLC, a Columbia University spin-off that develops life science software and provides data science consulting services.

Dr. Tatyana Goldberg is postdoctoral researcher who applies machine learning algorithms to answer various biological questions. In particular she is interested on creating computational methods that predict where proteins express in the living cell. Tatyana also focuses on understanding micro‐world warfare, that is predicting and identifying which bacteria can lead to infectious diseases. Tatyana is an ELES scholar who holds the prestigious Dimitris N. Chorafas award for excellence in research. She is also an experienced lecturer, who has been successfully leading students through scientific projects and international competitions such as “The CAFA Challenge” and “Google Summer of Code.”

======

Title: Video Recommendations: Find what you were (not) looking for

Author: Dr. Anca Tudoran

Abstract: The multimedia community is in a race to provide the best recommendation systems and navigation experiences for viewers. Various competitions, such as TRECVid and MediaEval, have emerged to foster this growing interest of the multimedia community. One of the tasks proposed at these competitions is Video Hyperlinking. The focus of this task is on improving the user navigation experience in a large video collection by offering information seeking and browsing capabilities in addition to search. The idea is that of creating links that originate from parts of video content and point to other relevant content leading to potential serendipitous encounters. The links can be seen as recommendations for potential viewers whose intent is not known at the time of linking.

This talk will present approaches for bringing diversity in links in order to improve the chance that any viewer will find at least one interesting link to follow. The increase in diversity will leverage several models among which bimodal topic models, based on speech and images.

Bio: Anca is now a data engineer in the Data Technology team at ProSiebenSat.1 Media SE. She has done a PhD at INRIA Rennes, in the field of video content analysis, focused on semantic structuring of video content from speech and video hyperlinking.