16th Belgium NLP Meetup

Sometimes it's nice to go back to where things all began. For the 16th Belgium NLP meetup we'll return to the location of our very first event in 2016, when Bert was just a character on Sesame Street and transformers were nothing more than toys for kids. The room will be larger, but the formula is still the same: three compelling speakers discuss what's hot in NLP. Lisa Becker (ML6) will talk about the challenges of pseudonymization, Maxime De Bruyn (University of Antwerp) will tell us whether language models know what they don't know, and Tim Van de Cruys (KU Leuven) will explore how creative they can be. Doors will open at 7pm and talks will start around 7.30pm. Afterwards we'll go and grab a drink at a bar nearby.

Below are the two confirmed abstracts for the evening. I'll add the third one soon.

Know What You Don't Know
Maxime De Bruyn, University of Antwerp
Language Models gained incredible popularity and size in the past few years. For example, the largest dense model, PaLM, possesses 540 billion parameters: sixteen times more parameters than all words on Wikipedia. This increased storage capacity allows large language models to display excellent common sense and world knowledge. However, they cannot know everything. This presentation explores whether large language models know what they do not know. During the presentation, you will experiment with large language models' internal knowledge by playing a live contest of Twenty Questions, so don't forget your smartphone.

The importance and pitfalls of pseudonymization
Lisa Becker, ML6
A lot of data we work with contains personal or other sensitive information. Thus we need to anonymize or pseudonymize the data to protect the identity and privacy of individuals. In this talk Lisa will explain the difference between anonymization and pseudonymization, give us all a short reminder of how the GDPR is involved, why data protection is so difficult and why there is no one-size-fits-all approach. The talk will be rounded up with a few examples of how Skyhaus/ML6 has handled and pseudonymized/anonymized sensitive data.

Shall AI Compare Thee to a Summer's Day?
Tim Van de Cruys, KU Leuven
Today's language processing systems are exclusively task-based:
typically, a machine learning model is trained on a significant amount
of training data, optimizing its parameters in order to produce the best
possible output for a particular problem or task. As such, the bulk of
natural language processing algorithms merely mimic human language use, and seldom produce creative output. In this talk, we'll explore whether
neural NLP models can be used for creative applications, such as poetry
generation and translation. Paradoxically, we will do so by imposing
constraints on the network architecture, forcing it to look for novel
ways to perform its task.