[Prédire les élections avec du deep learning] Quentin Chenevier

[qchenevier]

Le speaker

Quentin Chenevier - Lead Data Scientist & Product Manager @airbus

• Github: @qchenevier
• Linkedin: @qchenevier
• Twitter: @q_chenevier

Sujet de votre conférence

Predicting the Elections with Deep Learning - experiment in 2 parts:

• Results
• Modeling: Training a model on aggregated & anonymized labels (blog post coming soon)

Description de votre conférence

Results:

• No I didn't find a way to predict the elections results
• But It's surprising how much you can learn about voters behavior (who votes for which party), even using only aggregated public data. --> cherry picking of some interesting results
• It's only qualitative results.
• Still, it's interesting (worrying?) to think about what would be possible with more data (e.g: FAANG)

Modeling:

• A few schemas about how to code in pytorch a model to train on aggregated & weighted data.
• Explainability modules used to get the results (SHAP)

Informations diverses

• Thématique, Labels : Deep learning / Explainability / Social sciences
• Niveau de difficulté (débutant|avancé|confirmé) : débutant
• Durée: 30 min
• Format: slides
• Dispo: dans 1 mois