PyData 2020 January
Details
We kick off 2020 with a new venue in the heart of Kreuzkölln, two great talks and of course great food, drinks and lots of fun.
PLEASE NOTE: We are setting the attendee limit purposefully very high. Seats at the meetup will be on first come first serve basis. We have space for about 100 people.
Talks:
Gosia Adamczyk and Gunar Maiwald
Making image classification effortless with Image ATM (Automated Tagging Machine)
At idealo.de we store and display millions of images. Our gallery contains pictures of all sorts. You’ll find here vacuum cleaners, bike helmets as well as hotel rooms. Working with huge volume of images brings some challenges: How to organize the galleries? What exactly is in there? Do we actually need all of it?
To tackle these problems you first need to label all the pictures. Therefore we created Image ATM library - a one-click tool that automates the workflow of a typical image classification pipeline. With the help of transfer learning, Image ATM enables the user to train a Deep Learning model without knowledge or experience in the area of Computer Vision. All you need is data and spare couple of minutes! Our library tackles:
- Preprocessing and validating input images and labels
- Starting/terminating cloud instance with GPU support
- Training
- Model evaluation
In this talk we will briefly discuss the use-case that inspired us to create Image ATM. Next, we will jump into code.
Gosia currently works as a Machine Learning Engineer at Axel Springer AI - the artificial intelligence unit of Axel Springer SE. Previously she was a member of the Data Science Team at idealo.de.
Gunar Maiwald works as a Data Scientist with a strong focus on Software Engineering at idealo.de, an online price comparison portal based in Berlin-Kreuzberg.
-------
Branden Chan
Transfer Learning in Natural Language Processing (NLP)
In the last few years, the transfer learning paradigm has proven to be immensely effective and flexible in the field of NLP. Machine Learning based language models such as Google’s BERT now form the basis of almost every cutting edge NLP system, thanks to their ability to ingest massive amounts of unlabelled text and learn contextualised word embeddings. SOTA benchmarks have repeatedly been broken and previously challenging tasks such as SQuAD style question answering are all but solved. In this talk, we will look at the different components of a modern transfer learning system so that you can start training your own.
Branden Chan is a Machine Learning Engineer at deepset.ai, a Berlin based startup that is bringing cutting edge NLP techniques to industry. He is a Stanford Graduate in Computational Linguistics and is currently focused on training German language models to be used in question answering systems.
-------
Lightning talk by Simone Robutti on "Tech Workers' Coalition Berlin - A flash talk about who we are and what we are doing in Berlin for and with IT workers"