#34 Paris Women in Machine Learning & Data Science: joint meetup with Poznań
To continue our 2020-2021 European tour, we had a joint meetup with WIMLDS Poznań in May 2021.
👉 Coded bias, a documentary about a MIT Media Lab researcher Joy Buolamwini’s discovery that facial recognition does not see dark-skinned faces accurately, and her journey to push for the first-ever legislation in the U.S. to govern against bias in the algorithms that impact us all.
👉 The Mitchells vs The Machines, a family comedy movie where an eccentric family is saving the word against a robot uprising.
👉 The latest book by Kate Crawford, Atlas of AI, which gives perspective on the impact of AI on our lives, and how AI centralizes power.
The first speaker was Maria Ferlin, PhD candidate at the Gdansk University of Technology. Her presentation was entitled “Detect waste with AI ” and focused on the project “Waste detection in Pomerania”, which was a non-profit initiative to detect plastic waste in the environment.
The reason for such project is obvious: we create so much waste, and it sometimes lands in the wild. Even in treatment pipeline, it is useful to recognize the waste to help with recycling.
Several open datasets of waste exist, but they do not have the same classes, not always the same background… So to clean nature, Maria & her co-workers had to clean the data first. It’s still an unbalanced dataset, but at least more consistent. Pseudo-labeling is used to complete the dataset.
In the end, the results are quite good since the precision for detection is around 75%!
The second speaker was Oleksandra Vereschak, PhD Student in Human-AI Interaction (at ISIR, Sorbonne Université). Her presentation was entitled “How to Evaluate Trust in AI-Assisted Decision Making?”
Evaluating a systems with users is quite different from our usual performance metrics ;-) “Trust” is quite more complex to measure than model accuracy or blood pressure.
If you want to measure trust in a system, you need to remember at leas these three elements:
1. Trust cannot be directly determined by user behavior
2. Single-item questionnaires (“how do you trust the system on a scale of 1 to 7”) are limited
3. Multi-item questionnaires can be very long
And of course, trust may vary over time…
You can see here presentation below.
To know more about it, a paper of Oleksandra will be published soon in CSCW2021. Follow her on twitter to be posted @aleks_vereschak!
You can watch the entire meetup here:
It was our last event before our summer break. See you in September!
You can also:
📑 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.