mastodon
4.5.3
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
The warm fuzzy feeling when a former PhD student tells you they really like how your website is full of resources and they want to build something similar 🥰
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Alright, here are the slides:
https://cazencott.info/dotclear/public/publications/2026-05-21_azencott.pdf
And a thread to outline the main ideas.
The talk starts by briefly reminding what machine learning is: techniques that allow us to build models (which are functions, from a mathematical or computational points of view, that is to say, tools that apply a sequence of instructions to entries to produce an output) from data. In essence, you start from a template of what that sequence of instructions should look like; this template has a lot of parameters, that is to say, numerical parameters that you need to set. Learning means adjusting these parameters to fit the data.
1/6
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Me: "I'll be happy to give that talk, but due to my chronic condition, same as last time you invited me, it cannot be earlier than 11am."
Them: schedule my talk at 10am.
This and variations thereof happen so often I'm surprised when I *don't* have to go back and forth with organizers to obtain a later time slot.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
The claim that Anthropic's Mythos is so skilled at finding and exploiting software vulnerabilities that it's too dangerous to release has, unsurprisingly, nothing to do with security and everything to do with marketing.
"The flagship demonstration document turns out to be like the ending of the Wizard of Oz, a sorry disappointment about a model weaponizing two bugs that a different model found, in software the vendor had already patched, in a test environment with the browser sandbox and defense-in-depth mitigations stripped out. Anthropic failed, and somehow the story was flipped into a warning about its success."
The Boy That Cried Mythos: Verification is Collapsing Trust in Anthropic | flyingpenguin
https://www.flyingpenguin.com/the-boy-that-cried-mythos-verification-is-collapsing-trust-in-anthropic/
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
@ZikZak@oc.todon.fr @MonniauxD@social.sciences.re Pas du tout d'accord pour la première question. Les impacts environnementaux dépassent largement la mesure de la consommation (de quoi ? d'eau, d'électricité, de CO2, d'uranium ? par quoi ? une requête, l'entraîne d'un modèle, son développement ? on compte quoi dans son développement ? et comment on fait ça pour les plus gros acteurs, qui sont des boîtes privées puissantes qui se gardent bien de révéler quoi que ce soit à ce sujet ?), il faut compter aussi l'impact (matériaux, biodiversité) de la construction des centres de données et serveurs de calcul et des centrales qui les alimentent, celui de l'extraction des métaux rares permettant d'en construire les composants électroniques, etc. Je peux même pousser jusqu'à demander comment on prend en compte l'impact environnemental des guerres liées à l'accès à l'énergie ou aux terres rares.
Pour la deuxième, la question des logiciels n'est qu'une petite partie de la question de la souveraineté, celle des serveurs et centres de données en est une autre.
Quant à la troisième je ne sais pas trop ce qu'on peut y apporter niveau bac de philo, mais comme les précédentes, elle a des aspects techniques, réglementaires, sociaux, politiques, économiques, et ainsi de suite en plus de philosophes.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
La comm' de mon employeur me propose de faire une vidéo pour répondre aux questions suivantes :
- Comment mesurer et réduire l’empreinte carbone de l’IA ?
- Comment l’Europe peut-elle défendre sa souveraineté numérique ?
- Comment aller vers une IA responsable ?
...
en trois minutes 😁
La concision est une vertu que je ne possède pas en quantité suffisante pour cette mission.
(Évidemment je ne suis experte d'aucun de ces sujets, mais j'ai bien compris que si je ne m'y collais pas quelqu'un d'encore moins informé le ferait. Je pense fort à @MonniauxD@social.sciences.re pour la question de la souveraineté numérique, car je m'y connais encore moins que lui.)
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Reading about #mansplaining, I am reminded of that one time I was introduced to a tech transfer manager as a tenured member of a machine learning team. The guy then proceeded to explain machine learning to me. This was at a lunch, and half the people at the table stopped eating with their mouth half open.
And now I remember another one.
I was giving an intro to machine learning to an audience of biomedical researchers. One man had the following question: "I am wondering why you're calling it machine *learning*, when it obviously ought to be machine *teaching*. Maybe you've been confused translating from the French 'apprendre', as it can have both meanings? Have you checked the translation?" Yes, it had been said when I was introduced that I did my PhD in the US and that I was a machine learning expert. And then I shared that story on Twitter, and a man who was *not* a fluent French speaker told me that my story was probably wrong, because "apprendre" only means "to learn" and never "to teach".
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Had a blast today at Irisa! Very tiring day but what a warm welcome. There were so many attendees for my talk that we had to move rooms, and I got to sign off a copy of my textbook for a grateful student 😊 Many interesting scientific discussions as well, my head is buzzing.
Here are the slides of my talk on feature selection in high-dimensional genomics data: https://cazencott.info/index.php/pages/Talks
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Rappel :
Selon l'article 30 de la Loi pour une République Numérique, lorsqu'un article est le fruit de travaux financés au moins pour moitié par de l'argent public, ses auteurs et autrices ont le droit d'en mettre à disposition gratuitement en ligne la version finale acceptée pour publication (= celle qu'on produit, pas celle mise en page par l'éditeur), *même après avoir accordé des droits exclusifs à un éditeur*.
Seules réserves :
- il faut que les coauteurices soient d'accord
- si l'article est publié en Open Access, il faut (sauf accord de l'éditeur) attendre sa mise en ligne par l'éditeur
- sinon, un embargo de 6 mois en sciences, technique et médecine, 12 mois en SHS après la date de publication s'applique
https://www.legifrance.gouv.fr/jorf/article_jo/JORFARTI000033202841
Donc mettez vos papiers sur HAL s'ils ne sont pas en #openAccess
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
Professor at Mines Paris & Curie Institute. Machine learning for life sciences and therapeutic research. Shallow learning + fat data. #BiInSci. Cofounder of Paris WiMLDS. #MECFS patient.
