Martin Schuessler

Do Users Benefit From Interpretable Vision? Our user study shows that our baseline explanation technique outperformed concept explanations and performed similarly to invertible neural network explanations. Join us today at 7:30pm CEST iclr.cc/virtual/2022/p… #ICLR2022 @iclr_conf

I will run Deepseek-V3-Base 685B on M4 Mac Minis or die trying. 685B MoE with 256 experts -- perfect for Apple Silicon since they have a lot of GPU memory and only a small subset of params are active at once. Outperforms SOTA closed-source models on benchmarks including Aider.

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”

Remember when serious people said ridehail would help solve climate change? Oops. This new @ITS_UCDavis study found that most ridehail trips in California replaced transit (#1 substitute), biking, walking, or carpool or wouldn't have happened. doi.org/10.1177/036119…

Nicht die Fläche wählt, sondern die Menschen! Für die Darstellung wird jeder Landkreis in einen Punkt umgewandelt, wobei die Größe des Punktes proportional zur Anzahl der Wähler-/innen ist 👉 das Wahlergebnis wird dadurch besser abgebildet und weniger verzerrt.

Ist angesichts der Umtriebe der Prosecco-Nazis von #Sylt überfällig

Here is where #farmers storm the EU area at Maalbeek near EU HQ. "It looks like civil war" say EU officials as tractors overwhelm police. #FarmerProtest2024 #FarmersProtest2024

@overleaf Ok, still down. Git is also down. Now I need to work offline on an old copy... I am waiting for an offer from overleaf to make this right. A lost day of writing is very bad for all of us.

@overleaf It reserved this day to write on my thesis. I am also paying for overleaf. Consider waving my subscription fee for this month. This should not happen.

⚠️ Attention: ⚠️ There's an outage in one of our services and we are working to bring it back as soon as possible. Progress can be tracked via status.overleaf.com Sorry for the inconvenience it's causing.

During COVID, I collaborated with @martnsch and @oanaucs on how to evaluate interpretability in an online human subject study: arxiv.org/abs/2204.11642 We developed a rigorous study design, measured the true feature importance, and created a simple baseline.

@LeonSixt I totally agree! I have done an extensive literature review for user studies of explanation methods for image classification. It seems like there is no method that would beat our baseline. Either studies omit the baseline or the cant beat it.

- my impression is that XAI methods are in general too unreliable to be actually useful, - I would bet that the alignment of AI systems will be resolved without using XAI method

In a follow up work arxiv.org/abs/1912.09818, we found out that the positive matrices were to blame and that also other methods suffer from similar limitations. We recently updated the paper to fix a minor mistake in the proof and now also include a proof on the convergence speed.

I am very happy to celebrate my PhD defense. I want to take this opportunity to share my unique path through my PhD and share a small conclusion.

@ProfAndyField @sandeep3353 @Sage_Methods Can wait for the edition that uses tidyverse. I have inhaled the normal r version. Definitely shaped my PhD a lot. Now that I am in the industry I need the one book I can recommend to colleagues who want up their data analysis game. I hope this book will be it.

@sandeep3353 @Sage_Methods Next year in theory😬

Postman just delivered this, so I guess it’s out! @Sage_Methods

@ElonMuskAOC If XMail would be as good as Gmail in sorting email sure

Would you use 𝕏Mail over Gmail? Your data will never be collected, and your privacy will be protected.

Screw flying, this is what I want to see for the future 🤞🌙🚆 #Trainsbeforeplanes #Nighttrains

FB algorithms classify race, gender, and age of people in ad images and make delivery decisions based on the predictions. Images of Black folks are shown more to B users; images of children are delivered more to women; images of young women reach older men.

Trage Dich hier ein und solidarisiere Dich mit den Frauen im Iran! gruene.de/aktionen/hier-…

Funded and organised by @bifoldberlin, hosted at the #weizenbauminstitut: #BIFOLD Summer School 2022 - Ethics in Machine Learning & Data Management. About 30 Ph.D. students from the German Artificial Intelligence Network are here for lectures and workshops this week:

Another paper #ICML_2022: "A psychological theory of explanation" led by Scott Cheng-Hsin Yang and @Sjnjerak arxiv.org/abs/2205.08452 (version to be updated soon) Quick summary:

Watch our video presentation here: youtu.be/XSXwZFjZ4Aw

Do Users Benefit From Interpretable Vision? Our user study shows that our baseline explanation technique outperformed concept explanations and performed similarly to invertible neural network explanations. Join us today at 7:30pm CEST iclr.cc/virtual/2022/p… #ICLR2022 @iclr_conf

Work done by @LeonSixt @martnsch @oanaucs Philipp Weiß and @tlandgraf.