#35 Paris Women in Machine Learning & Data Science: satellite images, video categorisation & SEO Lesbienne
To start the 2021–2022 season, we have been lucky to be welcomed by Lifen on November 29th. It was our first in-person meetup in the last 22 months! In between, we had no less than 11 online meetups. We used that time to have joint meetups with european chapters: Italy, Germany, Poland… What a trip! We actually liked it, and plan to do some other online joint meetups this year: keep posted…
Going back to our 35th meetup, the attendees have been lucky to get, after so long, WiMLDS tee-shirts, sponsored by vpTech back in… September 2019. Many more are still waiting to be distributed: see you soon at our next in-person meetup 😉.
Marie Sacksick kicked off the meetup with some news our co-founder Chloé-Agathe Azencott will be there to introduce WiMLDS to the Paris neurips event. If you happen to be there too, come to say hello!
We are thinking about doing a scikit-learn sprint during the spring. Contact us if you are interested! 📩 firstname.lastname@example.org
According to our tradition, a host representative, Flavien Gilles, Lead Data Scientist at Lifen, presented the data science challenges they face in their everyday work. One of them is to identify entities in texts which can be medical prescription or medical invoices. This help to automatize the link between hospitals and MP, instead of going through postal exchange, which is slower.
In you are interested, you can contact Virginie Tramier, their Lead Talent Acquisition.
Alice Froideveaux, Lead Data Scientist at QuantCube Technology, opened the night with a talk on “How QuantCube Technology uses alternative data to create macroeconomic, financial and extra-fiancial indexes? Specific focus on the use of satellite images.” Her talk was deeply appreciated thanks to the technical elements she shared with us.
Alice started by presenting us what she calls alternative data: instead of using usual financial data from companies, it can using the number of ships in a port to estimate inport/export volumes. During that talk, she focused on finding cars on satellite images, in a project done with the CNES (Centre National d’Etudes Spatiales) and IRISA (Institut de Recherche en Informatique et Systèmes Aléatoires).
Among others, one interesting technical element is that with her team, Alice used data augmentation on prediction instead of training: each pixel has 20 predictions coming from the original image and 19 transformations, and a voting system is applied afterwards.
To know more about it, this work led to an article posted on arxiv.
Virginie Cornu, VP Data Jellysmack, took the second part to explain how categorisation on video was key for the company. She was impressive by how she managed both technical elements and business context.
Jellysmack detects and develops the world’s most talented video Creators through technology. The volume makes it mandatory to automate at least some aspects! Many machine learning fields are treated with the Data team: Natural Language Processing, Sentiment Analysis, Reinforcement Learning… One goal detailed in this presentation was: how can a video be classified and associated to a topic, e.g. animal video, reaction video, or DIY?
Interestingly, despite being a video classification task, the first version of the model was based only on text (title, tag, comments…), as it’s easy to collect, and easier to process than video. And the business results were good! Therefore, there was no need to complexify the models further.
Fanchon Mayaudon, cyberactivist and founder of SEOLesbienne, closed the evening by presenting how she launched this NGO back in 2019. Her presentation was full of emotion and she shared it magnificantly with the room.
In 2019 (and before), when googling “lesbienne”, the top 11 pages provided only porn. Not even a wikipedia page. This is not ideal for young girls wanting to discover their sexuality and role models, to say the least.
In June 2019 was celebrated Stonewall 50th anniversary. Around that time, two elements happened: it was the Ligue du LOL Affair, and #metoo. In that context, Fanchon decided to take action. There was only one meetup to decide an action plan to hack the word lesbian’s SEO. The movement was well broadcasted by the press, both in France and abroad.
Less than one year after the creation of the movement, Google made its first move to correct its algorithm!
This first victory against Google is to be celebrated! However, there are still some shady areas on the internet to fight: if you want to volunteer, you can join SEO Lesbienne on facebook and twitter.
If you want to keep posted about our activities or reach us, you are welcome to:
📑 check our Google spreadsheet if you want to speak 📣, host 💙, help 🌠
📍join our Slack channel for more discussions about machine learning, data science, and diversity in tech!
📩 send an email to the Paris WiMLDS team to keep in touch >email@example.com
🔥 Share your company or lab’s job positions for free on WiMLDS’ website.