We hope you had a great summer! We are very happy to announce that the VDSG is back from summer break, too, and our next meetup is happening in two weeks in A1!
This time we are having two talks: Segmentation in surveys and efficient machine learning for mobile devices.
Jelena Milosevic: “Efficient machine learning for mobile devices” (30-35 mins)
Increased amount of data allows for better training and more accurate machine learning systems. Big part of generated data today is coming from embedded and mobile devices, whose number is constantly on the rise. In order to fully profit from the collected information, artificial intelligence should come to these devices. However, this is currently difficult to achieve, mostly due to the computational demands of machine learning systems being too high for constrained computational resources of mobile devices. One of the reasons for this is that, when designing machine learning methods, most people only focus on accuracy, without taking into account constrained computational resources of developed solutions.
Many applications rely on efficient machine learning. Some examples are: vision and image processing, autonomous driving, and malware detection. In order to facilitate novel applications in these domains, it is of utmost importance to provide machine learning solutions that are not just accurate, but also suitable for constrained environments. In my talk I will discuss how we can design and develop such solutions suitable to be used in real-time, on-device, and at the same time customizable with respect to application requirements (accuracy, inference time, and power consumption).
Jelena Milosevic is passionate about machine learning and cybersecurity. She is currently a postdoctoral researcher at the Institute of Telecommunications, TU Wien, where she designs and develops machine-learning-based methods for detection of cyber attacks. She obtained her PhD in 2017 from Faculty of Informatics, University of Lugano, Switzerland, where her main focus was on the malware detection systems suitable for runtime usage on resource constrained systems and based on machine learning methods of low complexity. Previously, Jelena was an intern at IBM Cyber Security Center of Excellence in Beer Sheva, Israel, where she worked on the time-series analysis for anomaly detection and in Movidius an Intel Company in Dublin, Ireland, where she worked on the design and development of deep learning methods suitable for embedded environments. https://www.linkedin.com/in/milosevicjelena/
Marcin Kosiński: “Segmentation in Surveys using NMF” (30-35mins)
Working with high dimensional data? Often facing the need to group observations? This presentation is for you.
Segmentation should be balanced and distinctive, the discovered over- and under-indexed features within segments should create a meaningful story, and, ideally, the amount of differentiative factors that drives segmentation should be small.
The last requirement often becomes a bottleneck in a survey where respondents are asked an enormous amount of questions. One solution is the nonnegative matrix factorization that, in one attempt, segments respondents and their features! The concept of the NMF decomposition and applications in R will be presented with the explanation of diagnostic plots.
Marcin has a master degree in Mathematical Statistics and Data Analysis specialty. Challenges seeker and devoted R language enthusiast. In the past, keen on the field of large-scale online learning and various approaches to personalized news article recommendation. Community events host: organizer of Why R? conferences whyr.pl. Interested in R packages development and survival analysis models. Currently explores and improves methods for quantitative marketing analyses and global surveys at Gradient Metrics. https://www.linkedin.com/in/marcin-kosi%C5%84ski-81435aab/
Stay tuned and RSVP on our Meetup: https://www.meetup.com/Vienna-Data-Science-Group-Meetup/events/264362828/Slides are available here.You can find slides for this talk here.