19th Deep Learning Meetup in Vienna

Deep Learning is fueling more and more use cases in the AI domain. This time we have 2 interesting talks about how personalized product search is powered by Deep Learning:

Intro:
Visual Computing: then and now

Deep Learning had a tremendous impact on the visual computing domain. Error rates on image processing tasks dropped significantly and are still declining. In addition to this gain in accuracy, the complexity of the classification systems, as well as the skills and knowledge required, also decreased. In this short intro to our Visual Computing focused evening, Alexander Schindler will provide a brief introduction to traditional image processing approaches and put them in contrast to bleeding-edge image-labelling approaches to illustrate the simplification and elegance of Deep Learning models.

Talk 1:
Fast, Accurate And Customized Visual Similarity Search On Real-world Images

The first talk is given by Enes Deumić, Senior Data Scientist at Styria.ai and his colleague Vedran Vekić, Data Scientist at Styria.ai. They will present how they implemented visual search (search by image) on Willhaben.at, the biggest Austrian online second hand sales portal. To train their own classifiers using convolutional neural networks, data from the portal’s database was used, containing high amounts of noise and impurities. This data was not tagged but instead organized into categories incorporated into a hierarchy tree. The visual search system was built based on these classifiers. The entire system was built in stages including the users’ feedback in each stage. Due to that, the system significantly evolved over time, now empowering product search on Willhaben.at

Talk 2:
Mon Style - Machine Learning in the Fashion Domain

Our second talk is given by Matthias Hecker, CTO at Mon Style. Mon Style provides personalized recommendations in the fashion domain, thereby heavily relying on machine learning. The AI-fueled shopping assistant is able to get a precise understanding of the customer’s preferences and gives advice on what clothing styles, cuts, colors, prints and brands best fit. Mon Style will showcase their product classification pipeline for fashion detection and give some insights about their experience using mechanical turk for data annotation for visual recognition tasks in the fashion domain.

Hot Topics:
Rene Donner, Head of Machine Learning & Engineering at Contextflow will report from the International Conference on Learning Representations (ICLR), which took place in Vancouver in May, a high-profile conference with Yoshua Bengio and Yann LeCun as the General Chairs.

Join us for networking & discussions in the break and after the talks.
We are thankful for A1 Telekom to host this meetup in particular Stephan Wöber for the organization.

Looking forward to welcoming you.
Tom, Jan and Alex