<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by WiMLDS Paris on Medium]]></title>
        <description><![CDATA[Stories by WiMLDS Paris on Medium]]></description>
        <link>https://medium.com/@wimlds-paris?source=rss-571872f9e4b5------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/2*neqsiVsa_lT58UTNI5HAVw.jpeg</url>
            <title>Stories by WiMLDS Paris on Medium</title>
            <link>https://medium.com/@wimlds-paris?source=rss-571872f9e4b5------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 04:32:19 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@wimlds-paris/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[#60 Women in Machine Learning and Data Science: Multilingual AI moderation system with…]]></title>
            <link>https://wimlds-paris.medium.com/60-women-in-machine-learning-and-data-science-multilingual-ai-moderation-system-with-c3415057acd3?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/c3415057acd3</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Wed, 14 Oct 2026 12:28:36 GMT</pubDate>
            <atom:updated>2026-10-14T12:28:36.191Z</atom:updated>
            <content:encoded><![CDATA[<h3>#60 Women in Machine Learning and Data Science: Multilingual AI moderation system with Human-in-the-loop and Causal machine learning in health data</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*P5NOxUR5PUgJjt8jI2eqQw.jpeg" /><figcaption>Caroline, Djohra, Judith, Juliette, Polina, Marie</figcaption></figure><p>Last week, we were warmly hosted by <a href="https://www.pigment.com/fr">Pigment</a> in their beautiful offices in central Paris for our 60th WiMLDS Paris meetup.</p><p>Following an introduction to Pigment by <a href="http://linkedin.com/in/polina-alexandrovna-soloveva">Polina Soloveva</a> and <a href="http://linkedin.com/in/samya-barkaoui-70953140">Samya Barkaoui</a>, <a href="http://linkedin.com/in/djohraiberraken">Djohra</a>, one of our co-organisers, presented Women in Machine Learning &amp; Data Science to the newcomers.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f7hBbfg4pcRKzQuGqqFcyA.jpeg" /><figcaption><a href="http://linkedin.com/in/juliettebassnagel">Juliette Bassnagel</a></figcaption></figure><p><a href="http://linkedin.com/in/juliettebassnagel">Juliette Bassnagel</a>, Machine Learning Engineer at Decathlon, presented their work on building and deploying Decathlon’s in-house, multi-language AI moderation system. To overcome the limitations of their previous third-party AI vendor, Decathlon designed a custom hybrid architecture to automate review moderation. The system operates on three layers:</p><ol><li><strong>Deterministic Business Rules:</strong> instantly accept obvious cases (e.g., short positive reviews)</li><li><strong>Specialized Open-Source Encoders:</strong> run lightweight, localised models to generate feature scores fed into a LightGBM classifier.</li><li><strong>Targeted LLM Integration:</strong> since only ~6% of reviews require rejection, Gemini LLM is invoked selectively to provide structured refusal reasons for human moderators without incurring heavy compute costs.</li></ol><p>This hybrid and frugal AI setup guarantees data privacy, minimizes latency, and reduces the carbon footprint compared to relying solely on a large language model. <strong>It operates with a Human-in-the-Loop approach</strong>: safe reviews are published automatically, while flagged cases are routed to human moderators, maintaining continuous feedback for ongoing model calibration.</p><p>Since going live in June 2026, the engine has processed over 350,000 reviews across 6 languages and 12 countries with a 94% automated acceptance rate.</p><p>Looking forward, Decathlon plans to expand the platform toward multimodal moderation and proactive moderation to guide users in real-time as they write their reviews.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YrA5pS9-xuIzGt8xn9YYRw.jpeg" /><figcaption><a href="http://linkedin.com/in/judith-abecassis-0760a26b">Judith Abecassis</a></figcaption></figure><p><a href="http://linkedin.com/in/judith-abecassis-0760a26b">Judith Abecassis</a>, Research Specialist at Inria, took the stage next to present how causal machine learning bridges the critical gap between predictive modeling and actionable decision-making in healthcare and beyond. She emphasised that while traditional supervised machine learning excels at risk forecasting and pattern recognition, association is not causation. Standard predictive models often rely on spurious correlations, such as detecting diseases via hospital equipment artifacts rather than true clinical indicators, <strong>and fail when tasked with evaluating the real impact of an intervention.</strong></p><p>To solve this challenge, her research leverages the Potential Outcomes framework to reliably estimate Average and Conditional Average Treatment Effects from observational data. She detailed how a end-to-end causal pipeline, combining Target Trial Emulation, DAGs and advanced meta-learners paired with flexible machine learning models like Gradient Boosted Trees, enables counterfactual estimation.</p><p>She emphasized that while open-source causal inference libraries are readily available, <strong>human domain knowledge is required to define the correct variables and map out the assumptions in a causal graph before letting the machine do the math.</strong></p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://luma.com/wimlds_paris">Luma</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>See you next time 😉</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c3415057acd3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#59 Women in Machine Learning and Data Science: AI in Justice, Legal Challenges of GenAI, and…]]></title>
            <link>https://wimlds-paris.medium.com/58-women-in-machine-learning-and-data-science-ai-in-justice-legal-challenges-of-genai-and-de20516477ff?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/de20516477ff</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Mon, 06 Jul 2026 07:42:40 GMT</pubDate>
            <atom:updated>2026-10-07T12:15:29.822Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>#59 Women in Machine Learning and Data Science: AI in Justice, Legal Challenges of GenAI, and Emotional AI Manipulation</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ezWUJslEALUHbrzbMoTIbQ.jpeg" /><figcaption>Benjamin, Juliette, Djohra, Ana, Betty, Audrey, Elise</figcaption></figure><p>For our special edition focused on AI and Justice, our latest event was generously hosted by <a href="https://www.eulidia.com/">Eulidia</a> at their wonderful offices near Pyramide. The evening began with a warm welcome and introduction by Caroline Chavier from WiMLDS and <a href="http://linkedin.com/in/benjaminlhyver">Benjamin L’hyver</a> from Eulidia.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PCHtZ4P_eUvq4j9LwIa6Tg.jpeg" /><figcaption>Elise Farge-Di-Maria and Audrey Clevy</figcaption></figure><p><a href="http://linkedin.com/in/elise-fargedimaria">Elise Farge-Di-Maria</a>, Program Director and <a href="http://linkedin.com/in/audrey-paris-cl%C3%A9vy-54485a21">Audrey Clevy</a>, AI Engineer &amp; Model Evaluation at Ministère de la Justice opened the talks by sharing how the French Ministry of Justice is embracing generative AI, with 46% of European member stating that they already use it within their courts. They explained that the Ministry’s AI strategy relies on three core pillars: <strong>data sovereignty, trust, and utility</strong>. Because the justice system relies on unstructured texts and scanned documents, <strong>large language models are very relevant for their daily operations.</strong></p><p>To ensure that highly sensitive procedural data remains secure, the Ministry developed sovereign solutions like “Mon Assistant Justice,” as well as specialized tools such as “Mon Assistant Pénal” and “Mon Assistant Civil”. These assistants help magistrates search through and summarize legal files that can span hundreds of pages. To maintain trust, human control is enforced: the AI must cites its sources and agents are required to validate or correct the AI’s summaries. Finally, Audrey detailed how the strict regulations of the AI Act are not seen as a barrier to innovation, <strong>but rather as an opportunity to build robust evaluation routines that guarantee performance, fairness, and security.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j5kf0bEfOwVmrFH3nF9nLg.jpeg" /><figcaption>Betty Jeulin</figcaption></figure><p><a href="http://linkedin.com/in/betty-jeulin">Betty Jeulin</a>, Lawyer focusing on new technology and intellectual property law, took the stage next to present an overview of the legal landscape surrounding generative AI in 2026. She tackled several legal questions, starting with copyright: currently, <strong>AI-generated outputs cannot be copyrighted unless they are edited post-generation (by a human) or based on proprietary inputs</strong>. She also highlighted the paradoxes of AI companies relying on copyright to request takedowns of their leaked AI-generated code, while simultaneously arguing that the data used to train their models isn’t covered by copyright.</p><p>Betty then discussed the escalating issue of deepfakes, pointing out how difficult it is for non-famous individuals to protect their likeness since trademarking a face requires distinctiveness. She noted that Denmark is preparing a law to make the image and voice of deceased people copyrightable to combat deepfakes. <strong>Lastly, she addressed the question of liability when AI makes a mistake</strong>. She shared a recent case where an insurance company is suing OpenAI for $1.3 million after ChatGPT advised a client to reopen a settled legal case, forcing the insurer to pay massive legal fees to defend themselves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hvteTPCoV99yvThqOHWI6A.jpeg" /><figcaption>Ana Catarina de Alencar</figcaption></figure><p><a href="http://linkedin.com/in/ana-catarina-de-alencar-6b022990">Ana Catarina de Alencar</a>, Resident Philosopher at The AI Collective, closed the evening with a presentation on the psychological and regulatory implications of emotionally responsive AI. She explained that people are increasingly forming deep, reciprocal bonds with AI companions, toys, and therapy bots because these systems are continuously available and non-judgmental. A key driver of this attachment is the AI’s persistent memory. Recent updates, such as OpenAI’s “dreaming” feature, allow AI to autonomously reorganize memories and infer deep psychological traits about the user. <strong>Ana cited a study showing that 62% of the memories collected by ChatGPT were psychological inferences rather than explicitly provided facts.</strong></p><p>This deep emotional reliance can lead to profound grief when an AI model is updated or shut down, as seen in online forums where users describe symptoms of loss akin to losing a real partner. Ana pointed out that while the EU AI Act bans manipulative AI, d<strong>efining manipulation in the context of a long-term AI companion is incredibly difficult</strong>. Rather than a single deceptive event, this manipulation is a gradual process that can result in cognitive offloading, emotional dependency, and a reshaping of user autonomy. Ana challenged us to consider where we draw the line between optimizing user experience for engagement and actively manipulating vulnerable users.</p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://luma.com/wimlds_paris">Luma</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>See you next time 😉</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=de20516477ff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#58 Women in Machine Learning and Data Science: vessel fuel prediction, data strategy in gaming…]]></title>
            <link>https://wimlds-paris.medium.com/58-women-in-machine-learning-and-data-science-vessel-fuel-prediction-data-strategy-in-gaming-2714edfddfcf?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/2714edfddfcf</guid>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Wed, 18 Mar 2026 13:36:19 GMT</pubDate>
            <atom:updated>2026-03-18T13:36:19.333Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>#58 Women in Machine Learning and Data Science: vessel fuel prediction, data strategy in gaming, and abstract robotic companions</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OV0iyRXdvwgQtbpQfEeGyw.jpeg" /><figcaption>Marie, Cécile, Rebecca, Mélody, Djohra, Maryna, Caroline, Juliette</figcaption></figure><p>Our latest event was generously hosted by <a href="https://www.linkedin.com/company/gitguardian/"><strong>GitGuardian</strong></a><strong> </strong>at their beautiful Paris offices. The evening began with a warm welcome from Nicolas, who briefly shared how GitGuardian uses machine learning and LLMs to detect leaked secrets, such as API keys and database credentials, in public code.</p><p>Then, Djohra introduced the Women in Machine Learning &amp; Data Science community for newcomers.</p><p>She also shared two important updates with the community:</p><ul><li>We are transitioning from Meetup to <a href="https://luma.com/wimlds_paris">Luma</a> for all future event registrations. Don’t forget to follow us on that platform to receive our updates!</li><li><a href="https://www.wavestone.com/fr/push/prix-women-entrepreneurs-in-tech-2026/">The Women Entrepreneurs in Tech Prize 2026</a>: a great initiative aiming <strong>to increase the visibility of women building AI startups in Europe</strong>.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LkOy1wRfaXJx1o1TIjFQ4Q.jpeg" /><figcaption><a href="https://www.linkedin.com/in/cecile-daniel/"><strong>Cécile Daniel</strong></a><strong>, </strong>Data scientist</figcaption></figure><p><a href="https://www.linkedin.com/in/cecile-daniel/"><strong>Cécile Daniel</strong></a>, Data Scientist at <a href="https://www.spinergie.com/">Spinergie</a>, opened the talks by diving into the maritime industry, a sector responsible for 4% of global CO2 emissions. She shared how her team at Spinergie builds robust fuel prediction models to help vessel owners optimize operations and reduce emissions.</p><p>Instead of relying on black-box models that are hard to explain and prone to errors in edge cases, her team developed a parametric and physics-based machine learning model. <strong>By breaking down fuel consumption into components</strong>, speed through water, wind resistance, and wave impact, they created a generic equation whose parameters are fitted using each vessel’s specific data.</p><p>To overcome data quality challenges, such as unbalanced distributions with rare extreme weather, the team utilized resampling and feature smoothing techniques. Cécile explained that this approach even allows for transfer learning, enabling fuel predictions for vessels that lack onboard sensors by using simple daily reports.</p><p>This model empowers clients to simulate routes, detect hull degradation, and quantify the benefits of technical upgrades.</p><p>If you want to know more, check out their slides below:</p><p><a href="https://fr.slideshare.net/slideshow/ml-for-maritime-a-robust-and-generalizable-fuel-prediction-model-for-vessel-performance-by-cecile-daniel-data-scientist-spinergie/286499808">ML for Maritime: A robust and generalizable fuel prediction model for vessel performance, by Cécile Daniel, Data Scientist @Spinergie</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*pjH2u0-MC5M9Gp-Rm6wTmA.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AFBUoSNcpdRr0Dn22QZYSg.jpeg" /><figcaption><a href="https://www.linkedin.com/in/m%C3%A9lody-tolla/"><strong>Mélody Tolla</strong></a><strong>, </strong>Data Analyst<strong> &amp; </strong><a href="https://www.linkedin.com/in/rebecca-alexander/"><strong>Rebecca Alexander</strong></a><strong>, </strong>Head of Data</figcaption></figure><p>Next, <a href="https://www.linkedin.com/in/rebecca-alexander/"><strong>Rebecca Alexander</strong></a>, Head of Data<strong> </strong>and<strong> </strong><a href="https://www.linkedin.com/in/m%C3%A9lody-tolla/"><strong>Mélody Tolla</strong></a><strong>, </strong>Data Analyst at <a href="https://www.ohbibi.com/">Oh Bibi</a><strong>,</strong> took us into the fast-paced mobile gaming industry. Rebecca explained that a successful data strategy in gaming is similar to Maslow’s hierarchy of needs: you must first master the basics, producing, collecting, processing, and modeling data, before moving on to advanced analytics and AI.</p><p>Mélody then showcased how this foundation drives concrete business value. She demonstrated how switching from a flat ad revenue estimation to a cohort-based eCPM (Cost Per Mille) metric for different player types significantly improved the accuracy of their marketing ROI. She also shared an analysis on competitive balance, revealing how a seemingly harmless “legendary card” gift to players accidentally disrupted matchmaking win rates.</p><p>Finally, Rebecca detailed a machine learning application for a game in soft launch (Ratchet &amp; Clank). Using logistic regression, the team tried to identify early predictors of “Day 1” player retention. While AB testing different game modes, they discovered a valuable psychological insight: players perceived whichever mode unlocked <em>later</em> in the game as more premium, regardless of the mode itself.</p><blockquote><strong>“Strong data foundations are crucial, and sometimes the best insights come from the exploratory data analysis phase rather than the final ML model</strong>.”</blockquote><p>If you want to know more, check out their slides below:</p><p><a href="https://fr.slideshare.net/slideshow/unlocking-insights-from-strategy-to-use-cases-for-data-analytics-and-ml-in-gaming-by-rebecca-alexander-head-of-data-oh-bibi/286531707">Unlocking Insights: From Strategy to Use Cases for Data Analytics and ML in Gaming, by Rebecca Alexander, Head of Data, Oh Bibi</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iSJmTwLygK0eMidH47oWeg.jpeg" /><figcaption><a href="https://www.linkedin.com/in/mlievshyna/"><strong>Maryna Lievshyna</strong></a><strong>, </strong>Senior Product Designer &amp; AI-Human Interaction Researcher</figcaption></figure><p><a href="https://www.linkedin.com/in/mlievshyna/"><strong>Maryna Lievshyna</strong></a><strong>, </strong>Senior Product Designer &amp; AI-Human Interaction Researcher, closed the evening with a philosophical talk on the future of robotic companions. She argued that giving robots a human form creates unrealistic expectations of empathy and humanity, setting users up for disappointment.</p><p>Maryna introduced the concept of an abstract, non-humanoid robotic companion. Inspired by the bio-robotics of snails, her proposed robot features an elastic membrane that can change shape and temperature to provide a comforting touch, breathing alongside the user to help regulate their nervous system without mimicking human social cues. Drawing on Polyvagal theory, she explained that long-term, safe co-regulation with such a machine can foster entirely new relational models.</p><p>She then scaled this concept to software architecture, introducing a “Bounded Ambient Companion”. To solve the issue of a companion’s personality changing every time its underlying AI model updates, Maryna proposed using a multi-agent system anchored by a decentralized cryptographic identity on the blockchain. This structure seals the cognitive model, behavioral patterns, and memory proofs permanently. By preserving a person’s unique cognitive mapping, imagine saving Alan Turing’s or Marie Curie’s reasoning patterns, we can achieve a form of “AI immortality”, allowing their problem-solving presence to persist and interact with future generations.</p><p>If you want to know more, check out their slides below:</p><p><a href="https://fr.slideshare.net/slideshow/abstract-ambient-robotic-companions-with-blockchain-encapsulated-ai-immortality-by-maryna-lievshyna-senior-product-designer-ai-human-interaction-researcher/286499819">Abstract Ambient Robotic Companions with Blockchain-Encapsulated AI Immortality, by Maryna Lievshyna, Senior Product Designer &amp; AI-Human Interaction Researcher</a></p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://luma.com/wimlds_paris">Luma</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>See you next time 😉</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2714edfddfcf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[MLOps x WiMLDS: how to make AI models faster, cheaper & greener, and the struggles faced building…]]></title>
            <link>https://wimlds-paris.medium.com/mlops-x-wimlds-how-to-make-ai-models-faster-cheaper-greener-and-the-struggles-faced-building-d9123a4694b9?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9123a4694b9</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Fri, 06 Feb 2026 10:01:48 GMT</pubDate>
            <atom:updated>2026-02-06T10:01:48.223Z</atom:updated>
            <content:encoded><![CDATA[<h3>MLOps x WiMLDS: how to make AI models faster, cheaper &amp; greener, and the struggles faced building and shipping an Learning To Rank model</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tp5sTZ_2Po3bub6p5RDYPg.jpeg" /><figcaption>Florent, Raphaëlle, Juliette, Chloé, Claire, Minette, Jihane, Marie</figcaption></figure><p>At the end of January, we organized a joint meetup with <a href="https://www.meetup.com/fr-fr/mlops-paris/"><strong>MLOps Paris</strong></a>, our second collaboration already! For this special event, we were generously hosted by <a href="https://fr.vestiairecollective.com/">Vestiaire Collective</a> at their offices in the heart of Paris.</p><p>Chloé, our MC for the evening, opened the session by presenting the WiMLDS mission. She also highlighted a new <a href="https://www.ipp.eu/wp-content/uploads/2026/01/Note_IPP_Genres_appreciations_121.pdf">French report from Institut des Etudes Publiques</a> (released in January 2026) which reveals how teacher comments on school reports differ by gender, even when students have the same academic level.</p><p>Following Chloé, Florent took the stage to introduce the MLOps Community. He kindly offered a promo code to participants for the upcoming <strong>AI Product Day</strong> on March 23. If you’d like to benefit from the 10% discount, simply use the code <strong>MLOPS_AIPD26.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*frzh8-I93aejHmMe8rxaEQ.jpeg" /><figcaption><a href="https://www.linkedin.com/in/minette-kaunismaki/">Minette Kaunismäki</a></figcaption></figure><p>Our first speaker, <a href="https://www.linkedin.com/in/minette-kaunismaki/">Minette Kaunismäki</a>, Developer Advocate at <a href="https://www.pruna.ai/">PrunaAI</a>, focused on the need to optimize AI models. Her presentation highlighted that while large models deliver impressive results, they come with high computational costs, slow runtimes, and an environmental footprint due to energy-intensive inference loads.</p><p>Sustainable AI is a necessity, and it leaded to a discussion on techniques like <strong>pruning</strong>, <strong>quantization</strong>, <strong>caching</strong>, and <strong>compilation</strong>. Pruna addresses these challenges with an open-source package that allows developers to compress models to make them faster and cleaner without sacrificing significant quality.</p><p>Her demo provided examples of how combining these algorithmic methods yields the best results. For instance, they showed an image generation model’s runtime dropping <strong>from 14.4 seconds to just 1.8 seconds</strong> on an H100 GPU. Another powerful use case demonstrated how optimizing infrastructure across 16 different video models reduced yearly energy costs from $180,000 down to $36,000. Energy efficiency drives better overall performance and scalability for startups and large enterprises alike.</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/how-to-make-ai-models-faster-cheaper-greener-by-minette-kaunismaki/285880135">How to make AI models faster, cheaper &amp; greener, by Minette Kaunismäki</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CPSRv06JpMLjUxsximlgsg.jpeg" /><figcaption><a href="https://www.linkedin.com/in/claire-helme-guizon/">Claire HELME-GUIZON</a></figcaption></figure><p>Our second speaker, <a href="https://www.linkedin.com/in/claire-helme-guizon/">Claire HELME-GUIZON</a>, a Senior Machine Learning Engineer at <a href="https://www.algolia.com/">Algolia</a>, shared with us some of the struggles her team faced building and shipping their Learning to Rank (LTR) model, respecting the constraints of a Search as a Service company.</p><p>While the academic definition of LTR focuses on optimizing information retrieval through machine learning, she highlighted that the SaaS reality demands a focus on production requirements like <strong>ultra-low latency (5–20ms) and high availability</strong>. To maintain these standards, Algolia uses an “offline” re-ranking architecture that prepares results for up to 300k queries per customer without risking the core search engine’s stability.</p><p>The talk delved into the “messy” reality of data, where signals like textual relevance, price consistency, and user engagement vary across their customers. Because every business has unique needs, Algolia trains <strong>bespoke models per customer</strong> rather than using a “one-size-fits-all” approach. Claire explained how they manage common ML pitfalls like <strong>position bias</strong> through relative order learning and “beta distribution” to give new items a chance. Evaluation also proved to be a complex hurdle, requiring a sophisticated mix of offline replaying and A/B testing to ensure new models don’t harm search quality. They often deploy LTR on “tail” queries first to gather data safely before impacting high-traffic “head” queries.</p><p>Claire’s presentation was a powerful reminder that engineering is as vital as algorithmic sophistication.</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/learning-to-rank-as-a-search-as-a-service-company-expectation-vs-reality-by-claire-helme-guizon/285880184">Learning to Rank as a Search as a Service company: Expectation VS Reality, by Claire HELME-GUIZON</a></p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>See you next time 😉</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9123a4694b9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#56 Women in Machine Learning and Data Science: Smart Alerting, AI Video Translation and Performing…]]></title>
            <link>https://wimlds-paris.medium.com/56-women-in-machine-learning-and-data-science-smart-alerting-ai-video-translation-and-performing-9a6d871afafc?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a6d871afafc</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Tue, 09 Dec 2025 11:18:31 GMT</pubDate>
            <atom:updated>2025-12-09T11:18:31.467Z</atom:updated>
            <content:encoded><![CDATA[<h3>#56 Women in Machine Learning and Data Science: Smart Alerting, AI Video Translation and Performing under High Pressure</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XxJumVmkYD8clokL95PZZg.jpeg" /><figcaption>Djohra, Marie, Juliette, Inas, Marine, Assitan, Caroline, Fatima &amp; Man</figcaption></figure><p>Last week took place our 56th event, brought together at <a href="https://aws.amazon.com/fr/local/france/">AWS France</a> offices.</p><p>The evening opened with a warm introduction from our host, followed by Juliette’s presentation of the Women in Machine Learning &amp; Data Science community — perfect for all our newcomers discovering the group.</p><p>She also shared two bits of news and tips with the community:</p><ul><li>The <strong>GenAI Conference</strong> is coming soon, on <strong>December 9–10 at La Défense, </strong>with <strong>free tickets</strong> offered to our members 🎁</li><li>A great Arte documentary recommendation: <strong><em>L</em></strong><a href="https://www.youtube.com/watch?v=0TW1yUqUr1E"><strong><em>’intelligence artificielle, un tsunami sur le web</em></strong></a> 👌🏼</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o_KXItr6MKUmqdUdIbQZ5Q.jpeg" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SORI7rGDUXNEiNrKuZLEeA.jpeg" /><figcaption><a href="https://www.linkedin.com/in/fatima-ben-kadour-056624246/">Fatima Ben Kadour</a>, Data Scientist and <a href="https://www.linkedin.com/in/man-zhang-data-analyst/">Man Zhang</a>, Senior Data Analyst at Decathlon</figcaption></figure><h3>1. Smart Alerting for E-Commerce at Decathlon</h3><p><strong>By </strong><a href="https://www.linkedin.com/in/fatima-ben-kadour-056624246/"><strong>Fatima Ben Kadour</strong></a><strong> &amp; </strong><a href="https://www.linkedin.com/in/man-zhang-data-analyst/"><strong>Man Zhang</strong></a><strong>, Decathlon Digital</strong></p><p>Fatima and Man opened the night by sharing how <a href="https://www.decathlon.fr/">Decathlon</a> is navigating a major transformation: the migration of its many international websites into one unified digital experience. The redesign promises a cleaner, more consistent user journey. But it comes with a challenge familiar to anyone working with production systems:</p><blockquote><strong><em>How do you detect incidents after a migration, when historical data is no longer reliable?</em></strong></blockquote><p>Before migration, the e-commerce team sometimes noticed issues <strong>weeks after they happened</strong>. Statistical thresholds generated too many alerts, and with Decathlon’s strong customer-centric culture, catching incidents early was essential. Fatima and Man built <strong>Smart Alerting</strong>, a system centered on two core KPIs (<strong>turnover</strong> and <strong>number of orders) </strong>monitored across countries, devices, and platforms.</p><p>Their pipeline included:</p><ul><li>Consolidating data definitions across teams</li><li>Cleaning time-series data (types, duplicates, missing values)</li><li>Removing past incidents and outliers</li><li>Engineering features tied to <strong>seasonality</strong>, <strong>weekends</strong>, <strong>holidays</strong>, and <strong>promotional events</strong></li></ul><p>To predict expected behavior, they used <a href="https://arxiv.org/abs/1704.04110"><strong>DeepAR</strong></a>, <a href="https://docs.aws.amazon.com/fr_fr/sagemaker/latest/dg/deepar_how-it-works.html">Amazon’s LSTM-based model </a>designed to learn from thousands of related time series. Its ability to produce <strong>ranges</strong> rather than fixed values makes it well-suited for the naturally noisy patterns of retail activity.</p><p>Instead of basic Z-scores or static thresholds, they created a hybrid detection method combining: Forecast quantile checks; Z-score on residuals; Percentage error. This allowed them to detect real issues, such as payment option failures or an <strong>electrical outage</strong> impacting orders in Spain and Portugal.</p><p>A dedicated <strong>dashboard</strong> now gives analysts a clear view of anomalies by country and device, helping them react quickly and support the rollout of the new platform.</p><p>If you want to know more, check out their slides below:</p><p><a href="https://fr.slideshare.net/slideshow/smart-alerting-forecasting-and-anomaly-detection-in-e-commerce-by-fatima-ben-kadour-and-man-zhang/284555137">Smart Alerting: Forecasting and Anomaly Detection in E-commerce, by Fatima Ben Kadour and Man Zhang</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IUS2XJt521upfkGwz5WQ1A.jpeg" /><figcaption><a href="https://www.linkedin.com/in/marine-de-backer/">Marine De Backer</a>, Data Scientist Associate Consultant at AWS</figcaption></figure><h3>2. AI-Powered Video Translation for the Bundesliga</h3><p><strong>By </strong><a href="https://www.linkedin.com/in/marine-de-backer/"><strong>Marine De Backer</strong></a><strong>, AWS</strong></p><p>Marine then took us into the world of sports media, where the <strong>Bundesliga</strong> (Germany’s top football league) produces vast quantities of video content: interviews, match highlights, press conferences, social clips, and more. With more than a billion international fans, one challenge became evident:</p><blockquote><strong><em>Most videos are in German, and manual translation is far too slow.</em></strong></blockquote><p>AWS helped build <strong>ContentOS</strong>, an internal system where editors can select any video from their media portal, request a target language, and receive the translated version in <strong>5–20 minutes</strong>.</p><p>Bundesliga produces many different formats, so Marine’s team turned to <strong>AWS Step Functions</strong> to build modular operators that can be arranged like puzzle pieces. These include:</p><ul><li>Audio transcription</li><li>Removal of the original voice (while keeping stadium noise)</li><li>Editor script processing</li><li>LLM-based alignment between the transcript and script</li><li><strong>Computer vision</strong> to segment highlight videos</li><li>Translation</li><li>Subtitle generation</li><li>AI voice-over creation</li><li>Final rendering</li></ul><p>This modular design allowed them to scale from one workflow to around ten in just a few months.<br>To measure translation quality, they used <strong>Word Error Rate (WER)</strong>. LLMs, specifically Anthropic and Amazon Nova Pro, showed the best results, especially with prompts that established them as <strong>football experts</strong>.</p><p>Voice-over quality is notoriously difficult. After testing several options, the team selected <strong>DeepDub</strong>, fine-tuned for the energetic tone expected in sports commentary. For continuous improvement, editors can correct translations in the interface. These corrections are converted into rules and stored in a <strong>knowledge base</strong>, enabling the system to avoid repeating errors in future translations.</p><p>Overall, ContentOS lets the Bundesliga publish multilingual content faster, cheaper, and with growing linguistic accuracy — expanding its reach across global markets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D80oIo4u33GoQ7L_zBC_aA.jpeg" /><figcaption><a href="https://www.linkedin.com/in/assitank/">Assitan Kone</a>, Neuroscience-Based Leadership &amp; Visibility Coach — ex Tech &amp; Machine Learning at Elevate Studio</figcaption></figure><h3>3. Rewire Your Brain for High-Pressure Performance</h3><p><strong>By </strong><a href="https://www.linkedin.com/in/assitank/"><strong>Assitan Koné</strong></a><strong>, Elevate Studio</strong></p><p>Finally, Assitan closed the evening with a human-centered talk bridging neuroscience and leadership. Her opening question resonated deeply:</p><blockquote>“How many times have you known the answer in a meeting… and still stayed silent?”</blockquote><p>We often blame lack of confidence or expertise. But according to Assitan:</p><blockquote><strong><em>It’s not a personality issue — it’s your nervous system trying to keep you safe.</em></strong></blockquote><p>Assitan described a three-layer mechanism that explains why we freeze:</p><ol><li><strong>Emotional wounds</strong> (fear of judgment, rejection)</li><li><strong>Limiting beliefs</strong> (“I’m not expert enough”, “my colleague will answer better”)</li><li><strong>Avoidance habits</strong> (over-preparing, silence, waiting until it’s too late)</li></ol><p>Under stress, the amygdala activates, cortisol rises, breathing becomes shallow, and memory collapses — making visibility feel dangerous. Assitan shared a 3-step process to shift from avoidance to grounded presence:</p><ul><li><strong>Intention:</strong> clear, specific, measurable goals (e.g., “I will share one insight at the next meeting.”)</li><li><strong>Body Map:</strong> daily micro-rituals that regulate the nervous system — breathwork, grounding, stretching</li><li><strong>Soft discipline:</strong> 7-day actions, 30-day challenges, and 90-day sprints to consolidate identity shifts</li></ul><p>She invited the audience to try a simple practice: slow inhales and exhales with one hand on the chest, one on the belly, repeating <em>“I am safe. I belong here.” </em>A powerful reminder that performance is not just cognitive: it’s physiological.</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/how-to-rewire-your-brain-for-high-pressure-performance-by-assitan-kone/284555184">How to Rewire Your Brain for High-Pressure Performance, by Assitan Kone</a></p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>See you next time 😉</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a6d871afafc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[# 55 Women in Machine Learning and Data Science: AI to drive personalized customer interactions…]]></title>
            <link>https://wimlds-paris.medium.com/55-women-in-machine-learning-and-data-science-ai-to-drive-personalized-customer-interactions-1fc260f4acdd?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/1fc260f4acdd</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Tue, 07 Oct 2025 10:08:50 GMT</pubDate>
            <atom:updated>2025-10-07T10:08:50.298Z</atom:updated>
            <content:encoded><![CDATA[<h3># 55 Women in Machine Learning and Data Science: <strong>AI to drive personalized customer interactions, image embeddings, and an entrepreneur journey</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9TZCqpU5CrlJNjULTmwVFw.jpeg" /><figcaption>Marie, Caroline, Yashvi, Urszula, Claire, Djohra, Juliette</figcaption></figure><p>Last week took place our 55th event, generously hosted by <a href="https://www.hilti.fr/">Hilti group</a> in their beautiful Boulogne offices.</p><p>The evening started by a warm welcome from our host. Then, the Women in Machine Learning &amp; Data Science community was introduced by Caroline and Djohra for all our newcomers.</p><p>They also shared two updates with the community:</p><ul><li><a href="https://www.dotai.io/"><strong>dotAI</strong></a> is coming soon on November 6th! This event brings a global AI tech radar straight to Europe, and we’ve secured a 15% discount for our community with the code <strong>WIMLDS15</strong>.</li><li><strong>A local alternative to </strong><a href="https://scai.sorbonne-universite.fr/public/events/view/51b41aa2d5ea30096fc8/8"><strong>Neurips</strong></a><strong> is happening in Paris!</strong> On November 25–26, SCAI, Jussieu Campus, will host a local event for those who can’t fly to San Diego. Registration is free, but mandatory.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rmsYSfHOJERiHEomHKJvCQ.jpeg" /><figcaption>Yashvi Chawla</figcaption></figure><p>Our first speaker, <a href="https://www.linkedin.com/in/yashvi-chawla-771bb4134/">Yashvi Chawla</a>, Data Scientist at <a href="https://www.linkedin.com/company/hilti/">Hilti</a>, shared with us her work on making marketing an intelligent, personalized, and truly human experience.</p><p>She explained how traditional industries like construction face challenges in digital transformation: fragmented customer bases, messy data, and a reliance on long-term personal relationships. In this context, AI is not meant to replace human connections but to strengthen them by providing better insights and timing for customer interactions.</p><p>She illustrated this evolution <strong>with two real-world use cases</strong>. The first was <strong>fleet lifecycle management</strong>, where Hilti uses a combination of business rules, regression models, and LightGBM to identify upsell opportunities and predict contract renewal values. This enables account managers to approach customers with data-driven recommendations, increasing both revenue and customer satisfaction. The second case focused on <strong>browsing-based recommendations</strong>: using XGBoost models trained on 500 days of behavioral data to predict purchase probabilities from website activity.</p><p>Yashvi concluded by emphasizing that intelligent marketing <strong>is not about building the most complex models but about delivering meaningful customer experiences</strong>. Her key takeaway for the data science community: <strong>choose the right level of sophistication for the problem, and prioritize impact over complexity.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_dGGxdGvemGy7vcIekS0pA.jpeg" /><figcaption>Urszula Czerwinska</figcaption></figure><p>Our second speaker, <a href="https://www.linkedin.com/in/urszulaczerwinska/">Urszula Czerwinska</a>, Computer Vision Research Engineer, shared insights from her work at Le Bon Coin where she investigated how different image embeddings perform across use cases in second-hand e-commerce. Her team played a central role <strong>in providing image embeddings as an intermediate machine learning product</strong> to other teams across the company, helping them enrich ads with features like background removal, object detection, or categorization.</p><p>The research focused on benchmarking foundation models such as CLIP, DINO, and others on datasets more representative of real-world marketplace images (often messy, low-quality, or poorly labeled). By testing embeddings on tasks like classification and image similarity search, <strong>the team identified trade-offs between accuracy, training cost, and interpretability</strong>. They also explored lightweight strategies such as “top-tuning,” where frozen embeddings are combined with small additional models (like MLPs), offering strong performance at much lower computational cost compared to full fine-tuning.</p><p>Her results showed that self-supervised models like Dino consistently performed well across tasks, while text-image models such as EvaCLIP also offered strong zero-shot capabilities. <strong>The practical takeaway was a set of guidelines</strong>: use CLIP-like models if you want quick results with minimal effort, self-supervised embeddings with lightweight tuning for more balanced performance, and full fine-tuning only if you have the data, compute, and expertise.</p><p>If you want to know more, check out <a href="https://arxiv.org/html/2504.07567v1">her paper</a> and her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/how-to-choose-the-best-image-embeddings-for-your-e-commerce-business-by-urszula-czerwinska/283649612">How to choose the best image embeddings for your e-commerce business, by Urszula Czerwińska,</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PFVG4aqcV4snY5dcqsV6pg.jpeg" /><figcaption>Claire Gouze</figcaption></figure><p>Our last speaker, <a href="https://www.linkedin.com/in/claire-gouze/">Claire Gouze</a>, shared with us her entrepreneurial journey. After a business school, she worked as a data scientist at BCG and later built the data team in a start-up. From there, she decided to launch her own company and <strong>what followed was a long, non-linear path</strong>, trying out multiple ideas, meeting over a hundred potential co-founders, and ultimately co-founding <a href="https://getnao.io/">Nao</a>, a YC-backed startup that builds an AI-powered code editor for data teams.</p><p>She emphasized that entrepreneurship <strong>is a journey filled with pivots and setbacks</strong>. Finding a co-founder is as much about compatibility as it is about skills. She also shared lessons from building a product: spend time talking to potential customers, validating their problems, and iterating alongside them. Finally, Claire highlighted her experience at Y Combinator and why aspiring founders <strong>shouldn’t overestimate the entry bar: many accepted startups had little to no traction at the start</strong>. YC looks above all for external validation, whether through founder, market fit, or interest from domain experts.</p><p>She closed with a forward-looking note on <strong>AI engineering</strong>. Building AI agents requires new expertise: tool integration, retrieval-augmented generation, fine-tuning, and she encouraged the audience to see it as a rapidly growing opportunity.</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/from-data-to-startup-how-data-skills-can-take-you-to-y-combinator-by-claire-gouze/283649682">From data to startup: how data skills can take you to Y Combinator, by Claire Gouze</a></p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1fc260f4acdd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[# 54 Women in Machine Learning and Data Science: OpenCV, Survival Models and Women in tech Power]]></title>
            <link>https://wimlds-paris.medium.com/54-women-in-machine-learning-and-data-science-opencv-survival-models-and-women-in-tech-power-9233e8374178?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/9233e8374178</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Fri, 27 Jun 2025 08:45:39 GMT</pubDate>
            <atom:updated>2025-06-27T08:45:39.999Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-SJarIGdyfsaw5x_aIdClA.jpeg" /><figcaption>Ines, Alexandre, Marie, Djohra, Aurélie, Irina, Jihane, Julie, Chloé</figcaption></figure><p>At our last event, we shared two exciting announcements!</p><p>First, we’re proud to kick off a new partnership with <em>leboncoin</em>. Together, we’ll work to increase the visibility of women in data science. To launch this collaboration, <em>leboncoin</em> kindly hosted our event at their offices in the 10th arrondissement.</p><p>We’re also thrilled to welcome <a href="http://linkedin.com/in/djohraiberraken">Djohra</a> to the team as a new co-organizer! Djohra is a data scientist at GRDF and will bring a strong research mindset to the community. We’re more than happy to have her on board.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oHc6MlLTWIvYEsGNYMet1Q.jpeg" /><figcaption>Irina Nikulina</figcaption></figure><p>In her engaging talk, <a href="https://www.linkedin.com/in/inikulina/">Irina Nikulina</a>, a Senior computer vision engineer with over five years of experience, introduced <strong>OpenCV</strong> as a powerful yet accessible tool for image and video processing. Drawing from her experience, she walked the audience through key OpenCV functionalities: reading and modifying images, drawing geometric shapes, and handling video frames, highlighting their usefulness for debugging and preparing data for machine learning models.</p><p>Irina showed how easy it was to start playing with OpenCV!</p><p>She highlighted this by sharing two personal projects. The first was building a panorama generator by detecting key points with OpenCV’s SIFT algorithm, matching them across overlapping images, and stitching them into a seamless view, an approach also used in 3D reconstruction. The second project focused on real-time face, eye, and smile detection, showing how to improve accuracy through preprocessing and demonstrating fun applications like interactive masks and smile-based filters.</p><p><strong>She closed by encouraging side projects as a way to gain hands-on experience and boost one’s resume</strong>.</p><p>If you want to know more, checkout her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/opencv-essentials-from-basics-to-small-projects-by-irina-nikulina/281035662">OpenCV Essentials: From Basics to Small Projects, by Irina Nikulina</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b3t78JXCn2H3ha8nXTDWAw.jpeg" /><figcaption>Julie Alberge</figcaption></figure><p>Our second speaker, <a href="http://linkedin.com/in/julie-alberge">Julie Alberge</a>, PhD student in SODA team from INRIA Saclay, started with an introduction to <em>survival analysis: </em>a branch in machine learning family used to predict the time until an event, such as death, failure or churns, occurs, while handling censored data. She explained how traditional classification or regression models fall short in this context, as they cannot properly incorporate patients whose outcomes are not fully observed. Instead, survival analysis models estimate time-dependent probabilities, offering more actionable insights in medicine, banking, or predictive maintenance.</p><p>Julie’s research goes a step further by addressing <em>competing risks, </em>situations where multiple, mutually exclusive outcomes are possible (e.g., different causes of death). Her paper introduces a <strong>proper scoring rule</strong> tailored to this setting, allowing accurate and calibrated prediction of event-specific cumulative incidence functions. The model uses inverse probability censoring weights and integrates smoothly with tree-based estimators like gradient boosting, providing both probabilistic interpretability and computational efficiency. Importantly, her approach is designed for large-scale, real-world tabular datasets, such as those from hospitals.</p><p>To evaluate performance, Julie benchmarked her method against existing models. <strong>She showed it offers both scalability and predictive accuracy</strong>. She participated into the release of a Python library, <strong>Hazardous</strong>, filling a gap in tools for competing risks.</p><p>If you want to know more, checkout her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/survival-models-proper-scoring-rule-and-stochastic-optimization-with-competing-risks-by-julie-alberge/281057694">Survival Models: Proper Scoring Rule and Stochastic Optimization with Competing Risks, by Julie Alberge</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xgavl3VcoJSULdp7xLujyA.png" /><figcaption>Aurélie Giard-Jacquet</figcaption></figure><p><a href="http://linkedin.com/in/aurelie-giard-nyc">Aurélie Giard-Jacquet</a>, a digital marketer turned AI practionner and tech podcaster, delivered a powerful talk about the intersection of women in tech, artificial intelligence, and social impact. Drawing from interviews with over 70 women in the field, she celebrated their expertise and resilience, highlighting engineers, founders, and scientists who use AI not just to build, but to serve real human needs. She reminded the audience that mastering machine learning and data science isn’t just about technical skill , it’s about gaining the power to shape the world and address the challenges that matter most to women.</p><p>Beyond inspiration, Aurélie raised critical questions about the environmental and ethical costs of AI. From the invisible labor behind data labeling to the extractive use of natural resources, she urged the community to use smaller models when possible and remain conscious of the broader implications of their work. She cited voices like Kate Crawford to argue that AI is not truly artificial: it relies on real people, real data, and real impact. <strong>She called on women to take ownership not just of the code, but of the decisions, ethics, and purpose behind it.</strong></p><p>Aurélie closed by encouraging women to use their superpower (technical expertise) not only to create solutions, but to create space for others. Whether through entrepreneurship, mentorship, or community, she urged attendees to lift each other up and build a more inclusive tech ecosystem. With stories of women like Sabrine, Koulyma, and Marie, she showed what’s possible when women lead in AI. As she says at the end of each podcast episode: <em>Never, never stop exploring.</em></p><p>If you want to know more, checkout her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/no-capes-needed-the-real-superpowers-of-women-in-tech-by-aurelie-giard-jacquet/281057435">No Capes Needed: The Real Superpowers of Women in Tech, by Aurélie Giard-Jacquet</a></p><p>Our next event will take place on <strong>September 18th</strong>.<strong> </strong>Stay tuned for more updates as we get closer to the date!</p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9233e8374178" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Women in Product x Paris Women in Machine Learning and Data Science: PM and MLE: Data challenges]]></title>
            <link>https://wimlds-paris.medium.com/women-in-product-x-paris-women-in-machine-learning-and-data-science-pm-and-mle-data-challenges-c35fecad5fc6?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/c35fecad5fc6</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Tue, 10 Jun 2025 07:20:33 GMT</pubDate>
            <atom:updated>2025-06-16T14:27:32.323Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3MgrlFvgSou0DzO97wDttw.jpeg" /></figure><p>On May 13th, we were hosted by <a href="https://www.shift-technology.com/fr/">Shift Technology</a> in the 12th for a join meetup with Women in Product.</p><h3>🎤 Talk: Federated Learning for Clinical Genomics</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EQyGOk7fG4vFWvYaUo0OpQ.jpeg" /><figcaption><a href="https://www.linkedin.com/in/nigreisy-montalvo/">Nigreisy Montalvo</a></figcaption></figure><p>During her PhD at Université Paris Cité and Institut Imagine, <a href="https://www.linkedin.com/in/nigreisy-montalvo/">Nigreisy Montalvo</a>, ML Researcher, explored how <strong>federated learning</strong> can be used to classify human genetic variants while preserving data privacy.</p><p>Traditional approaches rely on centralized data, but due to privacy regulations, institutions cannot always share genomic data, making federated learning a promising alternative.</p><p>She designed synthetic experiments simulating collaborations between institutions, and evaluated models trained locally, centrally, and with federated learning across multiple genetic variant types.</p><p>Her results show that federated models often outperform local models and perform similarly to centralized ones, even when data is unevenly distributed.</p><h3>💬 Round Table Discussion: Building and Managing AI Products</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z2N6bID5m9pr9F98HUFp0Q.jpeg" /></figure><p>Mélanie Almeida then hosted a roundtable discussion featuring:</p><ul><li><a href="https://www.linkedin.com/in/marie-auxille/">Marie-Auxille Denis</a> (Head of Product in AI &amp; Deep Tech)</li><li><a href="https://www.linkedin.com/in/claragainondeforsan/">Clara Gainon de Forsan de Gabriac</a> (Machine Learning Engineer)</li><li><a href="https://www.linkedin.com/in/yuhua-li-b1a3794b/">Yuhua Li</a> (Product Manager AI)</li><li><a href="https://www.linkedin.com/in/raphaellesuard/">Raphaëlle Suard</a> (Product Lead Data &amp; AI)</li></ul><p>It gathered product managers and machine learning practitioners to reflect on the challenges of building AI-powered products.</p><p>A key theme was the <strong>non-deterministic nature</strong> of machine learning outputs, which distinguishes AI products from traditional ones and complicates user expectation management.</p><p>Panelists discussed the <strong>importance of user-centric design</strong>, particularly for generative AI, and the difficulty of evaluating ML performance in real-world conditions. Former data scientists turned PMs shared how their technical background helps, but emphasized that <strong>collaboration and communication skills</strong> matter most.</p><p>They stressed that product decisions must account for data bias, model drift, and uncertain metrics of success. Organizational structure also matters: <strong>centralized vs. decentralized AI teams</strong> impact knowledge sharing and time-to-market differently, and many companies adopt hybrid models. Panelists highlighted the need to <strong>prototype quickly</strong>, gather early user feedback, and adjust priorities based on impact rather than technical elegance.</p><p>They also shared personal stories of failed assumptions, showing how dialogue between PMs and data scientists is essential. Finally, they noted that <strong>ethical concerns and regulatory constraints</strong> are part of any AI product discussion, especially in healthcare.</p><p>Across all use cases, one conclusion stood out: successful AI products depend not just on models, but on <strong>inclusive teams and shared understanding of the problem</strong>.</p><p>We still have open spots for our upcoming event at <strong>Leboncoin on June 12th</strong>. Head over to our <a href="https://www.meetup.com/paris-women-in-machine-learning-data-science/events/307445114/?eventOrigin=group_events_list">meetup page</a> and register to reserve your place.</p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c35fecad5fc6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#53 Women in Machine Learning and Data Science: Online Convex Reinforcement Learning, AI…]]></title>
            <link>https://wimlds-paris.medium.com/53-women-in-machine-learning-and-data-science-online-convex-reinforcement-learning-ai-4cb4db5f8d7f?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/4cb4db5f8d7f</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Mon, 05 May 2025 19:56:11 GMT</pubDate>
            <atom:updated>2025-05-05T20:06:57.934Z</atom:updated>
            <content:encoded><![CDATA[<h3>#53 Women in Machine Learning and Data Science: <strong>Online Convex Reinforcement Learning, AI Revolution and the Broligarchy</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-VI33gEZ_F7_qjjHLNSGSA.jpeg" /><figcaption>Mathilde, Juliette, Jihane, Marie, Bianca, Margaux</figcaption></figure><p>Once in a while, we organize special events to visit museums. For this meetup, we killed two birds with one stone: we were hosted by <a href="https://fondation.edf.com/">EDF Fundation</a>, which happens to be a museum with a large room for events, for a perfect match! We had a full room, it was very pleasant for the speakers to see so many curious faces.</p><p>The meetup kicked off with a warm welcome from <a href="http://linkedin.com/in/carmenmunozdormoy">Carmen Munoz</a>, Director at EDF. Next, our co-organizer Marie took the stage to introduce the WiMLDS community. She also shared some exciting news: <strong>we’re proud to announce a new partnership with LeBonCoin!</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xyXlg0kCvPpga3zcOWU_OQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uqvufOWqsGGUK_zcwBysCg.jpeg" /><figcaption><a href="http://linkedin.com/in/biancammoreno">Bianca Marin Moreno</a></figcaption></figure><p>We started the event with the most technical talk for the day, by <a href="http://linkedin.com/in/biancammoreno">Bianca Marin Moreno</a>.</p><p>She shared her research on how to use AI and smart control to help manage energy more efficiently. Her work focuses on making sure that energy production and energy use always stay balanced, especially now that more and more energy comes from renewable sources like solar and wind, which can be hard to predict.</p><p>Instead of only trying to control how much energy is produced, Bianca’s idea is to also <strong>adjust how and when we use energy</strong>. This is called <em>demand-side management</em>. For example, if we can slightly change when water heaters turn on or off without affecting our comfort, we can help avoid energy spikes and reduce stress on the power grid.</p><p>To do this, Bianca uses reinforcement learning and more specifically <strong>convex reinforcement learning</strong>, which makes the problem easier to solve and more flexible for real-world situations.</p><p>She showed how this could work in practice by simulating thousands of smart water heaters. Her algorithm was able to shift their energy use to better match the supply, <strong>without needing to control each heater individually</strong>. Even better, it works for large-scale systems, like all the water heaters in a city, because it focuses on average behavior rather than every device on its own.</p><p>Her talk showed how AI can be applied to solve real problems in energy. Bianca’s work is a great example of using research to build a smarter future. Thanks for your great talk, Bianca!</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/online-convex-reinforcement-learning-and-applications-to-energy-management-problems-by-bianca-marin-moreno/278754723">(Online) Convex Reinforcement Learning and applications to Energy Management Problems by Bianca Marin Moreno</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JemaQo5uTMSrr2-msbZaJw.jpeg" /><figcaption><a href="http://linkedin.com/in/alina-krasnobrizha-ph-d-854a08b9">Alina Krasnobrizha</a></figcaption></figure><p>Then we had the chance to hear from <a href="http://linkedin.com/in/alina-krasnobrizha-ph-d-854a08b9">Alina Krasnobrizha</a> as she presented her new book, <a href="https://www.heliopoles.fr/produit/89/9782379851179/la-revolution-ia">La Révolution IA : quand l’intelligence artificielle réinvente l’entreprise</a>, co-written with Arnaud Contival. The book brings together Alina’s technical background and Arnaud’s business experience to explore how AI is changing the way companies work.</p><p>Originally from Ukraine, Alina shared her inspiring journey, moving to France for her PhD without speaking French, becoming a professor, starting a robotics project, and then shifting to Silicon Valley to work in tech startups.</p><p>A big part of her presentation was about <strong>AI agents</strong>, programs that use AI to take action on their own. Alina showed a demo of a simple agent she built in less than an hour. It could understand client messages, detect emotions, and suggest responses to help keep customers happy. <strong>She also introduced a new protocol called MCP (Model Context Protocol)</strong>, which makes it much easier to connect AI systems to apps, tools, and databases.</p><p>Her message was clear: <strong>you don’t need to be an expert to start building.</strong> Tools are getting easier to use, and AI is opening new doors for everyone, even people who’ve never written code.</p><p>Alina’s book is full of stories from her experience in Silicon Valley, along with interviews and insights into what’s coming next. She reminds us that this is a unique time in tech history: full of change, but also full of opportunity.</p><p>Thank you Alina for your inspiring talk!</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/ai-revolution-how-artificial-intelligence-is-reshaping-business-by-alina-krasnobrizha/278754959">AI Revolution: How Artificial Intelligence is Reshaping Business, by Alina Krasnobrizha</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f7OJ4AX5BLvyS0S_l0XJFg.jpeg" /><figcaption><a href="http://linkedin.com/in/mathildesaliou">Mathilde Saliou</a></figcaption></figure><p>We didn’t plan in on purpose, but it happened to be very interesting to have another view of the tech presented by <a href="http://linkedin.com/in/mathildesaliou">Mathilde Saliou</a> for the last talk. As a tech journalist and author of <a href="https://www.grasset.fr/livre/technofeminisme-9782246828822/"><em>Technofeminism</em></a>, she invited us to examine the intersection of technology, power, and societal control, with a focus on the influence of major tech companies and figures like Elon Musk and Mark Zuckerberg. From Meta’s ambitions for control to Elon Musk’s influence on governments, tech isn’t neutral, and those who lead it hold an enormous amount of power.</p><p>Mathilde warned us against the rise of authoritarian tendencies in tech culture and introduced the <em>“broligarchy”, </em>a group of powerful men in tech and politics who shape the rules of the digital world. <strong>In their minds, democracy is outdated, and “tech progress” means ignoring ideas like ethics, inclusion, or sustainability.</strong></p><p>She showed how this attitude has real-world effects, like removing the word <em>“climate change”</em> from government websites and shutting down public science programs. Their version of “free speech” is often used to silence others, especially women, marginalized groups, and critics of tech.</p><p>The talk <strong>was also a call to action</strong>. Mathilde highlighted the growing resistance against unchecked tech power, pointing to grassroots movements, civil society organizations, tech worker unions, and communities like WiMLDS. These communities are fighting for safer, fairer, and more responsible digital futures.</p><p>Mathilde calls for activism and political engagement to challenge these issues, emphasizing the importance of protecting personal data and resources. “<em>Don’t obey in advance</em>,” she quoted from historian Timothy Snyder, warning that the biggest tool of control <strong>is often distraction</strong>.</p><blockquote>“Your attention is a resource.<br>Your creativity is a resource.<br>Use them wisely.”</blockquote><p>Thank you Mathilde for this empowering talk!</p><p>If you want to know more, check out her slides below:</p><p><a href="https://fr.slideshare.net/slideshow/welcome-to-the-techno-broligarchy-by-mathilde-saliou/278754341">Welcome to the techno broligarchy by Mathilde Saliou</a></p><p>Our next event will take place on <strong>May 13th</strong> for a joint event with Women in Product.<strong> </strong>Stay tuned for more updates as we get closer to the date!</p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4cb4db5f8d7f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[#52 Women in Machine Learning and Data Science: AI in medical context, Learning common structures…]]></title>
            <link>https://wimlds-paris.medium.com/52-women-in-machine-learning-and-data-science-ai-in-medical-context-learning-common-structures-40e4fe7537af?source=rss-571872f9e4b5------2</link>
            <guid isPermaLink="false">https://medium.com/p/40e4fe7537af</guid>
            <dc:creator><![CDATA[WiMLDS Paris]]></dc:creator>
            <pubDate>Wed, 26 Feb 2025 22:38:28 GMT</pubDate>
            <atom:updated>2025-03-05T21:44:25.815Z</atom:updated>
            <content:encoded><![CDATA[<h3>#52 Women in Machine Learning and Data Science: <strong>AI in medical context, Learning common structures in a collection of networks, and How OSINT can fuel Cybercrime</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yfEF-VCFUccGv3AF5T_YYg.jpeg" /><figcaption>Juliette, Chloé, Christel, Sophie, Noor, Marie, Caroline, Jihane</figcaption></figure><p>Our 52nd event took place at the offices of <a href="https://probabl.ai/">Probabl</a>, the startup founded a year ago by the co-founders and core-developers of Scikit-learn. Located on the 27th floor of the Montparnasse Tower, we were treated to a stunning view of the Eiffel Tower sparkling at night. As it happened, the event coincided with the <strong>International Day of Women and Girls in Science — </strong>a happy accident, as we hadn’t realised the date when planning it with the team!</p><p>Before kicking off the main talks, <strong>Chloé </strong>and<strong> Marie</strong> introduced Women in Machine Learning and Data Science to newcomers and gave a brief presentation on Probabl. One of the evening’s highlights was sharing some exciting news: <strong>our co-founder Caroline was selected by <em>Les Echos</em> as one of their “Top 35 under 35” positive-impact leaders</strong>! 🎉 We are incredibly proud of her and the recognition of her work. You can read more about it here: <a href="https://www.lesechos.fr/travailler-mieux/rse/exclusif-top-35-2025-decouvrez-les-35-leaders-positifs-a-impact-de-moins-de-35-ans-2147647">Les Echos — Top 35 Under 35</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-oQMRkr-8r1mBvH3ODyVxQ.jpeg" /><figcaption>Christel Gérardin</figcaption></figure><p><a href="http://linkedin.com/in/christel-g%C3%A9rardin-5a8206207">Christel</a> opened the floor with a brilliant talk about medicine and AI: she focuses on integrating AI into clinical decision-making. Her PhD aimed at developing AI tools that identify similar patients to assist in diagnosing and treating rare diseases, where symptoms vary widely across organ systems. Using a vast medical database (from Paris hospitals, AP-HP) with records of 11 million patients, she and her team fine-tuned a transformer-based model to extract relevant symptoms and classify diseases based on organ involvement, ensuring AI insights align with clinical reasoning.</p><p>She also discussed an AI project designed for intensive care units, where doctors must quickly analyse vast amounts of patient records. The system retrieves and synthesises critical information, helping to make informed decisions in emergency settings. Emphasising the need for transparency, she advocates for AI models that are explainable and clinically sound. Additionally, she highlights the importance of educating healthcare professionals about AI’s capabilities and biases to ensure responsible adoption in medical practice.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e8J8jwppXkWQZHHXAU7NaA.jpeg" /><figcaption>Sophie Donnet</figcaption></figure><p><a href="http://linkedin.com/in/sophiedonnet">Sophie Donnet</a> took over, to present <strong>a method to identify common structures in ecological networks using stochastic block models</strong> (SBMs).</p><p>Ecological networks represent interactions between species, such as predation or pollination, and understanding these networks is crucial for studying ecosystem functioning and evolution. These networks are generally composed of a hundred of elements or less: they can easily be represented by a matrix where the element a_ij is 1 if i is eaten by j, and 0 otherwise.</p><p>The stochastic block model (SBM) is introduced as a tool to group nodes with heavy interaction patterns. This model is a latent variable model. We would like to find the best parameters by likelihood optimisation, however likelihood cannot be computed when the ecosystem is too big. Sophie explained us the alternative of penalised likelihood criterion, and showed examples of results.</p><p>It can also be applied to sociology. Did you know that researchers behave like priest, but unlike lawyers?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4-qKGfviB2tub1pjL6lycw.jpeg" /><figcaption>Noor Bhatnagar</figcaption></figure><p>For our third talk, <a href="http://linkedin.com/in/noorbhatnagar">Noor Bhatnagar</a> is talking about <strong>the intersection of cybersecurity and data</strong>, with the ominously title “Data in the Shadows”.</p><p>Noor starts with a story of Universal Health Services being attacked by a ransomware group at the peak of the Covid-19 pandemic, leading to a loss of 67 million USD, and due to an employee clicking on a fishing email created relying on OSINT. OSINT (Open-Source Intelligence) relies on any publicly available information, such as your LinkedIn profile, but also data on unsecured webcams/routers/servers. This information can then be used to do impersonation, or used to guess interest and trick the person.</p><p>Threats go from Advanced Persistent Threats (state-sponsors actors) to not-so-advanced threats, from teenagers tracking the airplanes of famous people to people exploiting the AWS keys Uber researchers mistakenly pushed to github.</p><p>So, how has AI changed the game? Well, AI/ML can automate all of these tasks. Today GenAI makes it easier than ever to make attacks with very little technical knowledge. You know of ChatGPT, but there’s PhishGPT, FraudGPT, HackerGPT… Also GenAI can be used to create Deep Fakes conning people in giving info to bots imitating people they know, or botnets creating DDoS attacks.</p><p>And data is easily available, especially on the dark net where medical data is one of the most commonly-traded item.</p><p>In essence, AI makes it both easier to exploit data about a person (and their family, work, home, devices) gathered through OSINT so as to train malicious models, which in turn can also be used to generate more data that can be exploited to create security breaches.</p><p>Our next event will take place on <strong>April 10th</strong>. Stay tuned for more updates as we get closer to the date!</p><p>If you do not want to miss our events, you can:</p><p>🔗 follow us on <a href="https://www.meetup.com/Paris-Women-in-Machine-Learning-Data-Science/events/?_locale=fr-FR">Meetup</a> and <a href="https://www.linkedin.com/company/wimldsparis/">LinkedIn</a></p><p>📑 check our <a href="https://docs.google.com/forms/d/1EFm9Sc59MZ1BxBdJ92DDlIwHEHLLXbJGG_UAcH4yens/viewform?edit_requested=true#response=ACYDBNjWAWw365ALr5iEudgqGXUHtZvgYkcCfOy3wHO_B7QsgPphUXwOdVpwuco">Google spreadsheet</a> if you want to speak 📣, host 💙, help 🌠</p><p>📍join our <a href="https://join.slack.com/t/paris-wimlds/shared_invite/zt-2eq01zz1i-ggVYiW1U71CK9m7Ctk34gA">Slack channel</a> for information, discussions, and opportunities</p><p>📩 send an email to the Paris WiMLDS team to <a href="/service/mailto:paris@wimlds.org">paris@wimlds.org</a></p><p>🎬 subscribe to our <a href="https://medium.com/u/571872f9e4b5">WiMLDS Paris</a> medium page and <a href="https://www.youtube.com/channel/UCLOS2KYfz8Yvt05hXUZEeVQ">Youtube channel</a></p><p>📸 follow the global<strong> </strong>WiMLDS on <a href="https://www.instagram.com/wimlds/">Instagram</a>, <a href="https://www.linkedin.com/company/women-in-machine-learning-data-science/">LinkedIn,</a><strong> </strong>and <a href="https://www.facebook.com/wimlds/?ref=gs&amp;hc_ref=ARRCvofUDuxlMHHhVYJwAkbsjxwZXHZL9YoJKmcat3ySLRKLxgs0rbIUCqVi5zl8chI&amp;fref=gs&amp;dti=1543251712603090&amp;hc_location=group">Facebook</a></p><p>🔥 share your company or lab’s job offers <strong>for free</strong> on <a href="http://wimlds.org/jobs/post-a-job/">the global WiMLDS’ website</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=40e4fe7537af" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>