Data Science Lightning Talks

REGISTRATION >> Please, in case you are attending register for this event.

We are thrilled to announce our next in-person event “DS Lightning Talks” on October 6th at Hotel Vincci Maritimo (sala Mar) in Barcelona, sponsored by Vista. Doors will open at 18:45h and talks will kick off at 19:00h. The security access control requires an attendee list.

With the aim of covering a wide range of real-life data science projects across different companies and institutes in the city, we will have four speakers giving lightning talks of 15 min each with 5 min of Q&A, followed by a networking session with refreshments.

Arnault Gombert (Head of Data Science at Citibeats)
Training ML models with low data volume
One of the main difficulties when you design a model is to find high-quality data that would make your model reliable in production. Sometimes existing literature provides well-known datasets that are perfect for the task you want to solve, but generally, your end-users’ needs can be slightly different from those addressed by existing state-of-the-art models: you may want your model to be multilingual (to scale worldwide), your classification model needs to go deeper in the classification (ranking of hate speech) or is totally tailor-made for your user (no clear literature on the task). We’ll see how we can leverage existing machine-learning literature on how to train a model with almost no data and reach decent results.

Baybars Kulebi (Project Lead - Senior Research Engineer at Barcelona Supercomputing Center)
Training 🤗transformers on ultra-hard difficulty
Most of the open DL frameworks out there are built to run on readily-available commercial hardware setups. However, when run on highly customized setups like High Performance Computing clusters, we might run into many different problems. This talk focuses on explaining 🤗transformers in detail, exploiting misadventures in HPC as a pedagogical tool.

Federica Capranico (Senior Data Scientist at Vista)
Advancing towards using ML on Vista’s Sustainability journey
Data Scientists are increasingly looking at ways to use “ML for Good.” At Vista, we are also pursuing our goals of increasing our own sustainability and that of our customers. In this talk, we’ll look at a single use case of how we can use ML to detect banners that will likely only be used one time and can therefore recommend more sustainable options to our customers.

Judit Xaubet (Senior Data Scientist at Novartis)
Forecasting with low quality data
At Novartis, we successfully developed a data science product that fulfilled the business needs when the data was scarce and sometimes inconsistent. This product is being used worldwide automatizing what was a manual forecast before and reducing the previous error by half.