Machine Learning PhD Student | Interpretability | FU Berlin | him/he

Berlin, Germany
Joined November 2016
I am about to miss the celebration for winning the Vesuvius Grandprize! This is the second time I am missing a Prize Ceremony for winning a Vesuvius Challenge in the USA and I started the visa application process as so as the prize was announced. This is incredibly frustrating and depressing, is there a way to expedite the process? The Ceremony is on March 16th. @usconsfrankfurt @usbotschaft @USAmbGermany I understand that you receive an overwhelming volume of applications and requests, but is there anything that could be done?
21
20
16
261
242,827
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.
1
1
3
1,179
- 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
3
1
4
898
Finally, the best part of my PhD was to collaborate with many amazing people: @tlandgraf, @Youssef_M_Nader, @martnsch, @oanaucs, @GranzMaximilian, @karlschulz_ and also during my internship at Google with @_beenkim and @martin_maas.
8
548
In the IBA paper, we also made another observation: the LRP's z+ rule fails the weight-randomization sanity check. Why does one of the most cited attribution methods return the same result even after the weights are partially randomized?
1
1
1
725
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.
3
3
1
47
5,890
We presented IBA as a method to explain classifiers. We were quite impressed to see how well the Information Bottleneck Attribution did in several empirical proxy tasks and received an ICLR Oral presentation for it: arxiv.org/abs/2001.00396
1
522
When starting my PhD, my original plan was to improve unsupervised learning using explainability methods. Our first paper with @karlschulz_ was supposed to be a step towards this direction, identifying the most relevant parts in an image using information theoretic principles.
1
507
Before starting my PhD, I already had my first academic success during my bachelors: arxiv.org/abs/1611.01331 Probably our paper was the first to apply GANs to a real-world problem, in this case the identification of honey bees.
1
1
503
It has been absolutely phenomenal working on the Vesuvius Challenge. The vision from @natfriedman and the foundation layed down by Prof. Seales and his team, make me every bit grateful to have the opportunity to score this touchdown!
Ten months ago, we launched the Vesuvius Challenge to solve the ancient problem of the Herculaneum Papyri, a library of scrolls that were flash-fried by the eruption of Mount Vesuvius in 79 AD. Today we are overjoyed to announce that our crazy project has succeeded. After 2000 years, we can finally read the scrolls: This image was produced by @Youssef_M_Nader, @LukeFarritor, and @JuliSchillij, who have now won the Vesuvius Challenge Grand Prize of $700,000. Congratulations!! These fifteen columns come from the very end of the first scroll we have been able to read and contain new text from the ancient world that has never been seen before. The author – probably Epicurean philosopher Philodemus – writes here about music, food, and how to enjoy life's pleasures. In the closing section, he throws shade at unnamed ideological adversaries – perhaps the stoics? – who "have nothing to say about pleasure, either in general or in particular." This year, the Vesuvius Challenge continues. The text that we revealed so far represents just 5% of one scroll. In 2024, our goal is to from reading a few passages of text to entire scrolls, and we're announcing a new $100,000 grand prize for the first team that is able to read at least 90% of all four scrolls that we have scanned. The scrolls stored in Naples that remain to be read represent more than 16 megabytes of ancient text. But the villa where the scrolls were found was only partially excavated, and scholars tell us that there may be thousands more scrolls underground. Our hope is that the success of the Vesuvius Challenge catalyzes the excavation of the villa, that the main library is discovered, and that whatever we find there rewrites history and inspires all of us. It's been a great joy to work on this strange and amazing project. Thanks to Brent Seales for laying the foundation for this work over so many years, thanks to the friends and Twitter users whose donations powered our effort, and thanks to the many contestants whose contributions have made the Vesuvius Challenge successful! Read more in our announcement: scrollprize.org/grandprize
15
26
1
363
38,499
Excited to see Jakob and Thore launch @PheironInc! Having had the pleasure of working with them, I highly recommend joining their team. If you're a ML Engineer, check out their latest job openings: pheiron.com/careers/senior-m… #MachineLearning #CareerOpportunities #MachineLearning
Over the past few months, @gogothorr and I have been heads down building @PheironInc . Our platform extracts disease markers from biomedical data to aim right: identifying the right targets for diseases and the right populations for clinical trials. This is personal to us:
2
4
705
Leon Sixt retweeted
Asking GPT-4 to think step-by-step and to reason about input validation and require statements can find the bug.😽 chat.openai.com/share/7b8af7…
Can ChatGPT audit smart contracts? Surprise: The answer is "No". Here's why:👇🧵
1
1
4
1,203
Our workshop continues with two very interesting spotlights 🔍 - on the gap between explanations and the explained model by @LeonSixt, and the use of Explainable AI for Scalable and Trustworthy AI by @coallaoh!
1
2
7
1,251