212 Followers
435 Following
176 Posts
mostly using bsky now: @janlause.bsky.social
data science postdoc in Tübingen 🧬🖥️🧠 scRNA data analysis, UMAP/tSNE & retina neuroscience | science journalism on AI & sustainability 🤖❤️🌍 | easily sidetracked by small plot details & cool birds 📈🔍🦜
Githubhttp://github.com/jlause
Websitehttps://hertie.ai/data-science/team/members/jan-lause
LocationTuebingen, Germany
Ich hab zuletzt als Kind eine Waldschnepfe gesehen. Seit ich den faszinierenden Artikel von @ThomasKrumenacker für @flugbegleiter redigiert habe, ist die Sehnsucht wieder groß. Dass 2021/22 in Deutschland (!) 10.000 Waldschnepfen legal geschossen werden durften, will mir zudem nicht so recht in den Kopf.
https://www.riffreporter.de/de/umwelt/voegel-vogel-portraet-waldschnepfe-ornithologie-vogelkunde-natur
Die Waldschnepfe – die große Unbekannte

Die Waldschnepfe ist wegen ihrer extrem heimlichen Lebensweise so etwas wie ein Geist des Waldes. Jetzt – im zeitigen Frühling – bestehen aber die besten Chancen, einen Blick auf diese ganz besondere Vogelart zu erhaschen.

RiffReporter

Resurrecting Recurrent Neural Networks for Long Sequences

Shows that careful design of deep RNNs performs on par with SSMs on long-range reasoning tasks with comparable speed.

https://arxiv.org/abs/2303.06349

Resurrecting Recurrent Neural Networks for Long Sequences

Recurrent Neural Networks (RNNs) offer fast inference on long sequences but are hard to optimize and slow to train. Deep state-space models (SSMs) have recently been shown to perform remarkably well on long sequence modeling tasks, and have the added benefits of fast parallelizable training and RNN-like fast inference. However, while SSMs are superficially similar to RNNs, there are important differences that make it unclear where their performance boost over RNNs comes from. In this paper, we show that careful design of deep RNNs using standard signal propagation arguments can recover the impressive performance of deep SSMs on long-range reasoning tasks, while also matching their training speed. To achieve this, we analyze and ablate a series of changes to standard RNNs including linearizing and diagonalizing the recurrence, using better parameterizations and initializations, and ensuring proper normalization of the forward pass. Our results provide new insights on the origins of the impressive performance of deep SSMs, while also introducing an RNN block called the Linear Recurrent Unit that matches both their performance on the Long Range Arena benchmark and their computational efficiency.

arXiv.org
RT @CharlieJGardner
My dear fellow conservation scientists - if the information we produce can't even be shown on TV these days, at what point will you agree that simply producing information is no longer enough? https://www.theguardian.com/media/2023/mar/10/david-attenborough-bbc-wild-isles-episode-rightwing-backlash-fears
BBC will not broadcast Attenborough episode over fear of ‘rightwing backlash’

Exclusive: Decision to make episode about natural destruction available only on iPlayer angers programme-makers

The Guardian
Ah yes, the good old "we tried chopping up the data many ways to find a specific three-way-interaction we had in mind". You love to see it in 2023.
Die schönste #aussenwerbung der Woche steht in #JenaLichtstadt am Westbahnhof.
Ich feiere das Haus der Weimarer Republik grad ein bisschen. ❤️
Statistical Rethinking week 10 has ended, and so has the course for this year. This week demonstrated analyses with allometry, state-based, and population dynamic models, and lots of general advice on planning, doing and reporting research. Memes of the week, explanations in alt text. Course website remains right here: https://github.com/rmcelreath/stat_rethinking_2023
GitHub - rmcelreath/stat_rethinking_2023: Statistical Rethinking Course for Jan-Mar 2023

Statistical Rethinking Course for Jan-Mar 2023. Contribute to rmcelreath/stat_rethinking_2023 development by creating an account on GitHub.

GitHub
Hello world! Elon Musk picking a twitter fight with Icelandic philanthropist and man of the year 2022 Haraldur Thorleifsson ended up being the final push that we needed, so now we are here on Mastodon. We don't know how any of this works, but follow us for posts on vision science, cognitive psychology, cognitive science, cognitive neuroscience, and stupid memes #VisionScience #CognitivePsychology #CognitiveScience #CognitiveNeuroscience #StupidMemes #ThanksForPointingOutHashtagsInComments

This is not a drill.

The nightmare scenario for personal privacy in a world where women are stripped of their reproductive rights is coming to pass:

Meta, Google, other tech companies are providing police with evidence to help prosecute women who seek or perform abortions

https://www.businessinsider.com/police-getting-help-social-media-to-prosecute-people-seeking-abortions-2023-2

Facebook, Google give police data to prosecute abortion seekers

Social-media sites are inundated with police requests for user data and may cooperate even if not legally required to, one legal expert told Insider.

Business Insider

Here is your must-read article for the day, a profile of @emilymbender, and her efforts to deflate the ridiculous hype around large language models such as ChatGPT.

It's also about the people who are behind that hype, and about what their way of thinking has the potential to do to us.

It's worth reading all the way to the end.

https://nymag.com/intelligencer/article/ai-artificial-intelligence-chatbots-emily-m-bender.html

‼️🫣Jetzt kommt es raus!

„Es wurde relativ schnell deutlich, es geht vor allem um Klimaschutz, nicht um Straßenblockaden.“

Hannovers Oberbürgermeister Belit Onay bei #Lanz über die Gespräche mit uns: