iiinova — where your digital self keeps evolving

Philipp Berens @philipp.hertie.ai · 28/06/2025

The colleagues from the new excellence cluster TERRA @unituebingen.bsky.social arriving after 100k on the 9th place!

120

Philipp Berens @philipp.hertie.ai · 26/05/2025

There should be a categorical imperative for bikes: Treat your bike like you would want to be treated by it

171

Philipp Berens @philipp.hertie.ai · 22/05/2025

This is where @hertie-ai.bsky.social will be located in three years or so!

150

Philipp Berens @philipp.hertie.ai · 22/05/2025

Importantly, we do not simply compute quality metrics, but perform a highly sensitive three-way odd-one-out study to assess the realism - even for ophthalmologists, it is hard to figure out, which images have been generated by our algorithm!

110

Philipp Berens @philipp.hertie.ai · 22/05/2025

This allows to generate high quality counterfactual images for individual patients, e.g. answering the question how an image had looked if the patient had been healthy or if she had been sick. The induced lesions look highly realistic, and the vessel structure is kept intact.

100

Philipp Berens @philipp.hertie.ai · 22/05/2025

In our model, we use the gradients of a robust disease classifier and regularization towards the original image to guide the diffusion process

110

Philipp Berens @philipp.hertie.ai · 22/05/2025

And another one: We show how to use a diffusion model for #fundus images for #counterfactual reasoning in #ophthalmology! If you ever wondered how the fundus images of your patient had looked, if he had been suffering from diabetic retinopathy, check this out ⬇️ journals.plos.org/digitalhealt...

1116

Philipp Berens @philipp.hertie.ai · 13/05/2025

This not only works much better than established post-hoc techniques, but also works well as a support tool for clinicians: Decisions for difficult cases improve and all decisions become about 20% faster!

100

Philipp Berens @philipp.hertie.ai · 13/05/2025

The explicit class evidence maps allows to penalize activations, e.g. via a sparseness penalty. This is extremely effective for diseases like #diabetic #retinopathy where lesions are small in initial disease stages: extracted high evidence regions almost always contain lesions.

100

Philipp Berens @philipp.hertie.ai · 13/05/2025

BagNets are modified ResNets with local receptive fields and an explicit class evidence maps at the end. We discussed there use for interpretable #medical #AI here: proceedings.mlr.press/v227/donteu2...

100

Philipp Berens @philipp.hertie.ai · 13/05/2025

We are excited that our work on inherently interpretable #deeplearning models for #AI for #medicine has been published in @plos.org #digitalhealth! You want to know how to combine the power of deep learning with accessible interpretation? This is for you! ⬇️ journals.plos.org/digitalhealt...

1184

Philipp Berens @philipp.hertie.ai · 03/05/2025

A new use case for ChatGPT: Make grants and other scientific texts adhere to US-style censoring.

190

Philipp Berens @philipp.hertie.ai · 03/05/2025

Ah, ChatGPT comes to your help:

000

Philipp Berens @philipp.hertie.ai · 14/03/2025

Are you looking for a #PhD or #Postdoc in #computational #neuroscience and #machinelearning for developing new models for testing theories of efficient coding? Join our team @hertie-ai.bsky.social at @unituebingen.bsky.social @ml4science.bsky.social! jobs.medizin.uni-tuebingen.de/Job/5996/Pos...

0129

Philipp Berens @philipp.hertie.ai · 14/03/2025

Are you looking for a #PhD or #Postdoc in #bioinformatics and #machinelearning for developing models for longitudinal #spatialtranscriptomics in the #retina? Join our team @hertie-ai.bsky.social at @unituebingen.bsky.social @ml4science.bsky.social! jobs.medizin.uni-tuebingen.de/Job/5995/Pos...

0125

Philipp Berens @philipp.hertie.ai · 08/11/2023

We are looking for an HPC Compute Cluster administrator to contribute to our ML Cloud! jobs.medizin.uni-tuebingen.de/Job/3574/HPC...

173