Can we improve neural prediction by incorporating efficient coding principles? Our study says yes! doi.org/10.1371/journa…. Many many great thanks to David Klindt Klaudia Szatko Dominic Gonschorek Lara Hoefling Timm Schubert Laura Busse Bethge Lab Thomas Euler .

Fun paper by teams of Bethge Lab & Samuel Albanie diving into the so-thought 'magic' zero-shot generalization properties of CLIP and Stable Diffusion-like models -- the authors study quite a lot of them (34 models and 5 pretraining datasets)
arxiv.org/abs/2404.04125

This year I decided to spice things up and move from an old English university town famous for ML, theology, and punting to an old German university town famous for ML, theology, and punting. Excited to be joining Bethge Lab as an ELLIS PhD student in machine learning!

are you a computer vision researcher or artist working on Neural Style Transfer? Bethge Lab Matthias Bethge
Justin Johnson Yongcheng Jing @Yezhou_Yang @ZoiRoup

check out pystiche, now officially 'Just published in JOSS' after going through review at pyOpenSci ( @[email protected] )

1/3

My Bethge Lab colleague @MaxGuenthner talking about modelling V1 with divisive normalization at #ccn19

Huge congratulations to my labs very first PhD 🎓🥰 🥇Dr. Steffen Schneider Steffen Schneider who’s also an ELLIS PhD w/Bethge Lab ❤️
NeurosciencePhD_EPFL M. Mathis Laboratory @EPFL

I’m so incredibly proud that now he starts his own lab at Helmholtz Munich | @HelmholtzMunich ➡️ he’s recruiting! dynamical-inference.ai

Is this the biggest heart you've ever seen? Extrapolating factors of variation from training is hard and even theoretically identifiable methods don't do well. Check out our new paper Amazon Web Services Bethge Lab Intelligent Systems

paper: arxiv.org/pdf/2107.08221…
code: github.com/bethgelab/InDo…

Excited for our panel at #NCMToy : Applications of deep learning in motor neuroscience !

W/ @tinpanhead Alexander Mathis Nidhi Seethapathi and myself, with Andrew Pruszynski moderating.

(CC: David Sussillo Krishna Shenoy Mackenzie Mathis, PhD Bethge Lab @UtopianCynic Kording —-& Lab 🦖 Kathleen Cullen)

Very interesting parallels between the nice new work from Bethge Lab and our work on RCN. arxiv.org/abs/1805.09190
1) Both emphasize the role of feedback, and generative models 2) Robustness vs maximizing accuracy on limited test set.

Heute war Bundesbildungsministerin Anja Karliczek 🇺🇦🇮🇱 am neuen Cyber Valley AI Research Building in Tübingen zu Besuch und hat ihre Unterstützung für europäische KI-Forschung betont. Vielen Dank für den Austausch! @TheresiaBauer Bernhard Schölkopf Universität Tübingen Bethge Lab Intelligent Systems BMBF

Deep networks are better than linear models at predicting visual cortical activity, but are also uninterpretable. Can you split the difference, and make better models that we can understand?

[Nice approach from Max Gunthner Bethge Lab Alex Ecker]
biorxiv.org/content/biorxi…

DNNs have a texture bias, but still use local shape when local features are informative (they have a local-over-global bias) journals.plos.org/ploscompbiol/a… Nicely fits with this work by Bethge Lab who show how you can overcome texture bias: blog.usejournal.com/why-deep-learn…

Richtig schön wars: Science Notes zur #KI in #Tuebingen . 💯 Danke an alle! Mehr Fotos: sciencenotes.de Wieland Brendel Bethge Lab Intelligent Systems Universität Tübingen Dr. Isabel Suditsch @selfOrgAnna CampusTV Tübingen @textboarder @snmagazin #CyberValley #KünstlicheIntelligenz #sciencenotes

“Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style” led by Julius von Kügelgen, Yash Sharma, and Luigi Gresele w/ Wieland Brendel, Bernhard Schölkopf, Michel Besserve (Dr.), and myself. Link: arxiv.org/abs/2106.04619 Intelligent Systems Bethge Lab Cambridge University [3/5]

Very insightful talk by Matthias Bethge at #BernsteinConference reviewing some of the striking differences between DNNs and the brain (adversarial examples, invariance...) and the solutions proposed at his lab Bethge Lab. (1/2)

The one TeaP@Home 2021 is in a symposium I put together with Katharina Dobs, featuring a smorgasbord of diverse DNN approaches in vision science, from Katharina and I, Bethge Lab, Martin Hebart and others not on Twitter.

Excited to see the launch of the NIPS 2018: Adversarial Vision Challenge on crowdAI! Thanks to a great collaboration with Bethge Lab Wieland Brendel B. Veliqi Jonas Rauber Sharada Mohanty @googleresearch @pennstate EPFL Nicolas Papernot Amazon Web Services crowdai.org/challenges/nip…

The call for satellite workshop at the #BernsteinConference is now open. Looking forward to your submissions. Deadline: May 10. bit.ly/BC21_workshops RT Marion Silies Jakob Macke Laura Busse TTchumatchenko Bethge Lab
Henning Sprekeler BCCN Berlin Laura B Naumann Philipp Berens sinzlab

We organized a joint workshop with Tinne Tuytelaars' lab at KU Leuven. We exchanged ideas on continual learning, object-centric representations, open set recognition and benchmarks. Well-spent two days in Cologne!