Home
Categories
EXPLORE
True Crime
Comedy
Sports
Society & Culture
Business
News
History
About Us
Contact Us
Copyright
© 2024 PodJoint
00:00 / 00:00
Sign in

or

Don't have an account?
Sign up
Forgot password
https://is1-ssl.mzstatic.com/image/thumb/Podcasts114/v4/6e/4c/f1/6e4cf16a-11d2-016c-fdd4-e145e15d049a/mza_10215802817563519116.jpeg/600x600bb.jpg
Post Mortem
François Paupier
26 episodes
1 week ago
In Post Mortem, engineers reflect on real-life incidents of IT systems they experienced. In each episode, we zoom on a specific event, ranging from a system outage, a cyber-attack, or a machine learning algorithm going wild with production data. We try to understand what happened and how the people behind those systems solved the situation. Along the way, you'll get hands-on advice shared by experienced practitioners that you can implement within your team to limit the risk of such incidents.
Show more...
Technology
RSS
All content for Post Mortem is the property of François Paupier and is served directly from their servers with no modification, redirects, or rehosting. The podcast is not affiliated with or endorsed by Podjoint in any way.
In Post Mortem, engineers reflect on real-life incidents of IT systems they experienced. In each episode, we zoom on a specific event, ranging from a system outage, a cyber-attack, or a machine learning algorithm going wild with production data. We try to understand what happened and how the people behind those systems solved the situation. Along the way, you'll get hands-on advice shared by experienced practitioners that you can implement within your team to limit the risk of such incidents.
Show more...
Technology
https://d3t3ozftmdmh3i.cloudfront.net/staging/podcast_uploaded_episode/8698156/2d088d128c82e5fa.jpeg
#8 When the facts change, I change my model
Post Mortem
23 minutes 28 seconds
4 years ago
#8 When the facts change, I change my model

"When the Facts Change, I Change My Mind. What Do You Do, Sir?" disait JM Keynes. 

L’économiste soulignait alors l’importance de réajuster ses a priori et sa représentation du monde lorsqu'on on est confronté à de nouveaux éléments.

C’est la même chose lorsqu’on entraîne un modèle de machine learning et qu’on le déploie.

Les données que l’on va rencontrer en production suivent-elles une distribution similaire aux données sur lesquelles on a entraîné le modèle? Si non, comment peut-on ajuster le tir?

Témoignage et retour d’expérience avec Hamza Sayah, Data Scientist @ Ponicode.


Références et concepts mentionnés


- Kullback-Leibler divergence, en un mot: une quantité qui mesure la dissimilarité de deux distributions de probabilités.


Pour une excellente vidéo donnant l’intuition derrière le lien entre l'entropie, l’entropie croisée et la KL divergence,  se référer à la vidéo d'Aurélien Géron "A Short Introduction to Entropy, Cross-Entropy and KL-Divergence" 

https://www.youtube.com/watch?v=ErfnhcEV1O8


- Pour l'intuition derrière le concept d'embedding, voir le blog post de Jay Alammar, "The Illustrated Word2Vec", https://jalammar.github.io/illustrated-word2vec/



- AST - Abstract Syntax Tree, https://en.wikipedia.org/wiki/Abstract_syntax_tree



- La recommandation de Hamza: le livre "La Formule Du Savoir", de Lê Nguyên Hoang


Post Mortem
In Post Mortem, engineers reflect on real-life incidents of IT systems they experienced. In each episode, we zoom on a specific event, ranging from a system outage, a cyber-attack, or a machine learning algorithm going wild with production data. We try to understand what happened and how the people behind those systems solved the situation. Along the way, you'll get hands-on advice shared by experienced practitioners that you can implement within your team to limit the risk of such incidents.