Tech Talks in Data Science, AI and Machine Learning
How do you implement information extraction from images and free text? Can you apply advanced machine learning techniques to real-world scenarios? Can a machine ever learn to identify the base-level in knowledge graphs?
On Tuesday 15th October, in collaboration with Databricks’s Spark+AI Summit and Women in Tech Netherlands, we hosted Tech Talks in Data Science, AI and Machine Learning to discover just this. With help from our Chair, Esther Schagen-van Luit, Data Science, ML and AI enthusiast, as well as being specialist master in security architecture at Deloitte, each talk was warmly received by the 400+ audience.
Clemence Burnichon – Depop
Speaking in front of a full-house at Amsterdam RAI, Clemence Burnichon, lead data scientist at Depop, started the evening with a talk on Understanding the Depop Inventory. Clemence is leading the machine learning effort across Depop, specialising in generating smart capabilities to improve buyer’s and seller’s experience with Depop.
In her talk, Clemence explained that at Depop, their 15 million+ users can list items for sale with up to 4 images and a short description. In order to understand their inventory, Clemence and her team have worked on extracting information from images and free text by developing a machine learning model and deploying there model in production. She walked us through the journey they took to create such capabilities from solution design to deployment in production.
Adi Polak – Microsoft
End-to-end data science solutions are a hard task. Applying advanced ML approaches to real-world scenarios requires us to clean, prepare and feature engineer the data before we can even start discussing the algorithms. But this is not all, we need to test various ML models, try different configurations and techniques and last but not least, productionize the models.
Doing so while dealing with massive amounts of data is not exactly an ‘easy walk in the park’.
In our second talk, Adi Polak, a senior software engineer and a developer advocate at Microsoft working on Azure, demonstrated basics of the ML development process. She showed the audience how to perform feature engineering with Apache Spark, what ML pipelines are, and the various options for training and managing ML models with Apache Spark.
In her role at Microsoft, Adi, focuses on microservices architecture, distributed systems, real-time processing, big data analysis, machine learning at scale and functional programming. As a developer advocate, Adi brings her vast experience in tech and helps teams to design, architect and build their software and infrastructure with cost-effective, scalability, team knowledge and business market in mind.
Laura Hollink – CWI
To close our Meetup, Laura Hollink gave her talk on How to Make Knowledge Graphs Work for Humans. Laura is a tenure track researcher at CWI with an interest in both knowledge representation and human computer interaction.
Laura explained how she enriches knowledge graphs by predicting which concepts are “basic level concepts”. The basic level, according to experiments in cognitive psychology, is the level of abstraction in a hierarchy of concepts at which humans perform tasks quicker and with greater accuracy than at other levels.
Laura developed a method to machine learn the basic level from data. A range of experiments, performed on WordNet, showed that basic level prediction works well with a representative training set. To predict the basic level in a new domain, a domain-normalisation step is necessary. Concepts that are difficult to label for humans seem to also be harder to classify automatically. Future plans are to apply, improve and test the method on a large scale and on a wide range of knowledge graphs. All this is required to help machines better anticipate human behaviour.
Did you miss this Meetup? Recordings of these presentations will be available soon.