We look back to the fifth edition with a BIG smile on our faces. What a turnout and we even got a great (w)rap-up by Siraj! Thank you all for helping us put Lisbon PyData on the map as one of the biggest data science communities in Portugal!
We can't wait for the sixth edition with talks of data scientists at Feedzai and James Finance. Read all about it below.
*** PROGRAM ***
18.40 - 18.55 | Walk-in
18.55 - 19.00 | Greetings and intro
19.00 - 19.40 | First talk incl. Q&A - "Counterfactual analysis: working around the impact of machine decisions" by Feedzai
Many machine learning systems take actions that have direct influence in the real world. Examples of this are credit scoring systems that approve or decline loans, or fraud detection systems that block transactions that seem suspicious. All of these systems suffer from a common problem: when we act to prevent unwanted outcomes, we never get to know whether those outcomes were really going to happen. Were they True Positives or False Positives? How can we then estimate the performance of our classifiers? And how can we use these events to train models later on?
This talk will be about these issues and how we tackle them in the fraud detection use case at Feedzai. We'll show how ignoring the problem can have nasty consequences, go through strategies to estimate classifier performance, and introduce techniques to train classifiers using (some) unlabelled events.
Mariana Almeida and Pedro Abreu are research data scientists working at Feedzai, applying machine learning techniques to stop fraud in financial transactions. Before joining Feedzai, Mariana Almeida worked for 4 years on natural language processing after doing a PhD and a Postdoc on image processing and Pedro Abreu worked as a robotics research engineer.
19.40 - 20.10 | Break with drinks & snacks
20.10 - 20.20 | Community pitches
Community members are welcome to introduce themselves and share their experiences. If you would like to give a lightning pitch please get in touch.
20.20 - 21.00 | Second talk incl. Q&A - "Whitebox machine learning: explainability through model agnostic approaches" by James Finance
Human regulators in heavily regulated fields like credit scoring require mission critical predictive systems to be inspectionable and interpretable in order to avoid problems with problematic features usage & interactions. For a long time, organizations in these fields were restricted to simple & interpretable models or restricted to use just one type of non-parametric models (e.g.: tree ensembles) in order to allow the regulator (and management) to inspect the behavior of the model.
With new laws coming into place, like Global Data Protection Regulation (GDPR), organizations face even more restrictions to the development of predictive systems. As such, this can be a "make or break" moment for machine learning & predictive systems as a whole. Fortunately, new model agnostic methods have been developed that allow organizations to cope with these issues and, also, open a range of new possibilities for interacting/dialoging with their predictive systems.
In this talk, Bruno will introduce you to the new & hot field of machine learning explainability/interpretability. We will take a hands-on approach, showing you how you can implement model-agnostic methods to inspect your model.
Bruno Dias is one of the data scientist within James R&D team. He has a background in computer science (intelligent systems & robotics) and has been working on machine learning explainability and missing data analysis/imputation. Also, he has experience with front-end development, audio signal processing and semantic web. Also, he has a big interest in healthcare informatics and computational creativity.
21.00 - 21.30 | Drinks and snacks
Looking forward to seeing you there! If you would like to present at a future event, don't hesitate to reach out to us.
Lots of love,
The Jungle Crew
Claim the event and start manage its content.I am the organizer