Wat we doen
When you spend your days in the company of Bert, Ernie, Elmo, Roberta, Albert and their kin, sometimes you long for some real company. Luckily the Belgium NLP Meetup is there to help you find some real human contact. Our 12th meetup will take place on Tuesday December 3rd, in the offices of AI company ML6 in Ghent. As always, our doors open at 7pm, talks start around 7.30pm, and from 9pm there's time for networking and drinks.
Here's the program for the evening:
It's alive! Designing and improving real-world NLP for production systems
Thomas Vrancken & Anna Krogager, ML6
In this talk, ML6 will share some insights they obtained during the development and go-live of a recent NLP-powered client application. What worked and what didn't? The application in question is a search engine based on document embeddings. They will discuss how to host their ML models on google cloud, integrating the search engine with elasticsearch, and how they use kubeflow pipelines to automate their workflows.
Embedding Rare Words: how (not) to get the most out of sparse data
Jeroen Van Hautte, TechWolf
Distributional semantic models have become the bread and butter of the NLP scene: take a look at any recent language processing system and you will probably find Word2Vec (or similar) under the hood! While immensely useful, these models struggle when facing rare words, which is problematic in applications with niche documents or professional language. Jeroen will talk about his research on this topic, as well as how it helps TechWolf build better products.
Information Extraction from Medical Notes for Early Birth Risk Prediction
Lucas Sterckx, UGent-imec
Preterm birth is, with a worldwide incidence of 5 to 18%, a major contributor to neonatal morbidity and mortality. In order to prevent under- and overtreatment, it is important to differentiate patients who will deliver prematurely and those who will not. By combining clinical observations and doctors’ free text notes, machine learning models can predict the chance of giving birth within the next seven days. This way, the therapy can be adjusted accordingly to optimize outcome for mother and child. Processing medical notes during hospitalization is a critically important application area of natural language processing, for which there are few robust, practical, publicly available models. Clinical narratives have unique characteristics that differentiate it from scientific biomedical literature and the general domain, requiring a focused effort around methodologies within the clinical NLP field. Lucas will discuss the NLP pipeline developed by IDLab and Ghent University Hospital to bring structure to medical notes.