Johannes Roth

Machine learning engineer & PhD researcher.

At the Max Planck Institute in Leipzig, I work on vision models and the brain. Before that, I did machine learning at ScaDS.AI and CHECK24.

Explore my work Get in touch

Available from January 2027

Johannes Roth smiling, in a light blue Oxford shirt with rolled sleeves and his arms crossed

Selected work

Projects

GitHub

Horrorverse

A horror film recommender with mood search, an explorable map of the genre and an illustrated history.

Open Horrorverse
A contact sheet of natural photographs from the dataset, one highlighted

ReLAION-natural

Naturalness scores for every image in ReLAION-2B, plus CLIP embeddings for the natural photos.

View dataset

LAION-fMRI

A deeply sampled fMRI dataset of natural images.

Explore project

re:vision

A community challenge testing whether findings in visual neuroscience replicate on LAION-fMRI.

Explore project

Conwell replication

A replication on LAION-fMRI of a large study on which design choices make vision models brain-like.

View on GitHub

thingsvision

An open-source library for extracting and comparing features from vision models.

View library

Publications

Google Scholar

Open-source tools

GitHub
  • voice-annotate

    A macOS app: talk and draw on your screen, then paste the result into any coding agent.

  • slurmboard

    A lightweight dashboard for Slurm jobs, logs and resource use on HPC clusters.

  • fmriqc

    An interactive QA and review tool for fMRI data snapshots.

  • cvmanova_python

    Cross-validated MANOVA for multivariate fMRI pattern analysis.

  • pysearchlight

    Customisable searchlight analysis for Python.

  • how-to-sample

    Code and experiments for the CCN 2025 paper on sampling natural images.

Writing

All writing

  • A century of fear

    What scares us changes. An illustrated analysis of 19,302 horror films, from atomic mutants to masked killers to ghosts.

  • It's always clear from the outside

    A passage from Art and Fear and what it says about how we see other people's work versus our own.

Experience & education

Full CV

Timeline

Education

  • May 2022–present

    Doctoral candidate · Dr. rer. nat.

    University of Giessen

    Research focus: machine learning and computational neuroscience.

  • 2017–2021

    M.Sc. Computer Science

    Leipzig University · Grade 1.2

    Master’s thesis: generative deep learning for visual processing in the human brain (grade 1.1).

  • 2014–2017

    B.Sc. Business Information Systems

    Leipzig University · Grade 1.5

Roles

  • May 2022–present

    PhD Researcher · Computational Cognitive Neuroscience

    Max Planck Institute for Human Cognitive and Brain Sciences & University of Giessen

    Independent research on vision models and human brain activity, with GPU-based data pipelines, fMRI experiments, and public research software.

  • Jun 2021–May 2022

    Research Scientist · Machine Learning in Medicine

    ScaDS.AI Dresden/Leipzig

    Deep-learning models for brain-tumor segmentation and predictive models for clinical data, including mortality prediction and uncertainty estimation.

  • Nov 2019–May 2021

    Data Scientist / ML Engineer · Working student

    CHECK24, Travel Vertical

    Designed and deployed a production computer-vision service for 20 million hotel images. Improved recommendation-data pipelines, reducing update time by over 95%.

  • Oct 2020–May 2021

    Full-Stack Developer · Freelance

    Kimetric UG

    Delivered Django websites for Hebart Lab and the THINGS Initiative, from client requirements to deployment, Linux hosting, and maintenance.

  • Oct 2018–Sep 2019

    Data Scientist · Working student

    Webdata Solutions GmbH (now Vistex)

    Developed machine-learning models for product matching and image classification in e-commerce, raising image-based matching accuracy from 12% to 83%.

  • Feb 2018–Aug 2018

    Data Analyst · Working student

    Mercateo Services GmbH & Co. KG

    Data integration, reporting, and testing for an AWS-based data warehouse using Apache NiFi and Pentaho.

Life outside work

Climbing, making music, books, and dinner with friends.

Meet the other side of me
Johannes bouldering on a rock face in a mountain valley
Movement
Johannes holding an electric guitar
Music
A colorful plate of food on a wooden table
Food
A quiet forest path near the meditation hall
Meditation

Still Johannes.

A little less formal.

Here's what life looks like away from the screen: climbing, making music, reading, and having friends over for dinner.

A bit about me Back to work
Johannes smiling in a T-shirt and climbing harness

A bit about me

Johannes bouldering on a rock face in a mountain valley

Movement

My love for moving my body started with joining a breakdance crew at 17. These days I mostly boulder, lift weights, and do yoga. A day in the mountains with some climbing is hard to beat.

Johannes holding an electric guitar

Music

A few years ago, I asked some friends if they wanted to start a band. Everyone said yes. We mostly play pop punk covers at birthday parties, but the dream of writing our own songs is still alive.

A colorful plate of food on a wooden table

Food

I used to be an incredibly picky eater. Now I love throwing whole food ingredients together and having friends over for dinner. Hosting is a blast, even when it gets surprisingly stressful.

A quiet forest path near the meditation hall

Meditation

Meditation and I have had an on-off relationship for over a decade. A ten-day silent retreat in 2024 changed a lot for me. I walked this forest path probably a hundred times and kept discovering something new.

Music

I listen to a bit of everything. Spotify probably knows my taste better than I do.

My Discover Weekly

Horror movie night

My wife and I watch horror movies with friends. I built a little recommender to help us choose.

Find our next movie My Letterboxd

Writing

All writing

  • A century of fear

    What scares us changes. An illustrated analysis of 19,302 horror films, from atomic mutants to masked killers to ghosts.

  • It's always clear from the outside

    A passage from Art and Fear and what it says about how we see other people's work versus our own.

On my shelves

81 loved books

Reading now

  • Art & Fear by David Bayles & Ted Orland
  • The Mind Illuminated by Culadasa

Loved

  • Silence by Thich Nhat Hanh
  • Old Path White Clouds by Thich Nhat Hanh
  • You Are Here by Thich Nhat Hanh
  • No Mud, No Lotus by Thich Nhat Hanh
  • Peace Is Every Step by Thich Nhat Hanh
  • When Things Fall Apart by Pema Chodron
  • Full Catastrophe Living by Jon Kabat-Zinn
  • Wherever You Go... by Jon Kabat-Zinn
  • Why We Meditate by Daniel Goleman & Tsoknyi Rinpoche
  • The Art of Living by William Hart
  • Be Here Now by Ram Dass
  • After the Ecstasy, the Laundry by Jack Kornfield
  • The Attention Revolution by B. Alan Wallace
  • The Book: On the Taboo by Alan Watts
  • The Wisdom of Insecurity by Alan Watts
  • Autobiography of a Yogi by Paramahansa Yogananda
  • Doors of Perception by Aldous Huxley
  • The Psychedelic Explorer's Guide by James Fadiman
  • Food of the Gods by Terence McKenna
  • Red Rising series by Pierce Brown
  • The Kingkiller Chronicle by Patrick Rothfuss
  • A Song of Ice and Fire series by George R.R. Martin
  • The Expanse series by James S.A. Corey
  • The Belgariad series by David Eddings
  • Mistborn series by Brandon Sanderson
  • The 13th Paladin series by Torsten Weitze
  • The Three-Body Problem by Cixin Liu
  • Dune by Frank Herbert
  • It by Stephen King
  • Island by Aldous Huxley
  • Malice by John Gwynne
  • Of Blood and Fire by Ryan Cahill
  • Gods of the Wyrdwood by R.J. Barker
  • Magician by Raymond E. Feist
  • The Dwarves by Markus Heitz
  • Uldart: Die dunkle Zeit by Markus Heitz
  • Der Turm der Bettler
  • A Wizard of Earthsea by Ursula K. Le Guin
  • Ender's Game by Orson Scott Card
  • Braiding Sweetgrass by Robin Wall Kimmerer
  • Wild by Cheryl Strayed
  • Into the Wild by Jon Krakauer
  • Into Thin Air by Jon Krakauer
  • Seven Years in Tibet by Heinrich Harrer
  • Born to Run by Christopher McDougall
  • A Search in Secret India by Paul Brunton
  • Adult Children of Emotionally Immature Parents by Lindsay C. Gibson
  • How to Do the Work by Nicole LePera
  • Why Has Nobody Told Me This Before? by Julie Smith
  • Why Zebras Don't Get Ulcers by Robert M. Sapolsky
  • Maybe You Should Talk to Someone by Lori Gottlieb
  • In the Realm of Hungry Ghosts by Gabor Mate
  • How to Change Your Mind by Michael Pollan
  • DARE by Barry McDonagh
  • Feeling Good by David D. Burns
  • A Liberated Mind by Steven C. Hayes
  • Four Thousand Weeks by Oliver Burkeman
  • Deep Work by Cal Newport
  • The Myth of Normal by Gabor Mate
  • How Not to Die by Michael Greger
  • How Not to Age by Michael Greger
  • Food Rules by Michael Pollan
  • Omnivore's Dilemma by Michael Pollan
  • Deep Medicine by Eric Topol
  • Total Immersion by Terry Laughlin
  • 24/6 by Tiffany Shlain
  • Smartphone Nation by Kaitlyn Regehr
  • It Doesn't Have to Be Crazy at Work by Jason Fried & David Heinemeier Hansson
  • The Age of Surveillance Capitalism by Shoshana Zuboff
  • How to Break Up with Your Phone by Catherine Price
  • Make Time by Jake Knapp & John Zeratsky
  • We Should All Be Feminists by Chimamanda Ngozi Adichie
  • Men Explain Things to Me by Rebecca Solnit
  • The Road to Reality by Roger Penrose
  • Surely You're Joking, Mr. Feynman! by Richard P. Feynman
  • Tools of Titans by Tim Ferriss
  • How to Win Friends and Influence People by Dale Carnegie
  • The Life-Changing Magic of Tidying Up by Marie Kondo
  • When Breath Becomes Air by Paul Kalanithi
  • Man's Search for Meaning by Viktor Frankl
  • That Good Night by Sunita Puri
Books I found okay
  • The Obstacle Is the Way cover

    The Obstacle Is the Way

    Ryan Holiday

  • Stolen Focus cover

    Stolen Focus

    Johann Hari

  • The Wim Hof Method cover

    The Wim Hof Method

    Wim Hof

  • Atomic Habits cover

    Atomic Habits

    James Clear

  • Indistractable cover

    Indistractable

    Nir Eyal

  • Waking Up cover

    Waking Up

    Sam Harris

  • Built to Move cover

    Built to Move

    Kelly Starrett

  • Kitchen Confidential cover

    Kitchen Confidential

    Anthony Bourdain

  • The Comfort Crisis cover

    The Comfort Crisis

    Michael Easter

  • The Shallows cover

    The Shallows

    Nicholas Carr

Horrorverse

A horror film recommender with mood search, an explorable map of the genre and an illustrated history.

What it is

It started as a way to settle what to watch on our horror movie nights. Describe a mood, like slow-burn folk horror under 100 minutes, and it picks one to three films for the night. Rate a few films and the daily picks get better, or wander the atlas: about 20,700 films arranged by theme and story.

What I did

I built the recommender, trained on MovieLens 32M with a content bridge for films nobody has rated yet, plus the mood search on text embeddings, a movie-night chat where an LLM turns what you say into search settings (it never does the ranking), the atlas, Letterboxd import, accounts and the self-hosted deployment.

What came out of it

It's live, and we use it every movie night. Its catalogue also became the data behind A century of fear, an illustrated history of 19,302 horror films, each labelled by two Claude models and checked against Wikidata.

Open HorrorverseMovie night
A contact sheet of natural photographs from the dataset, one highlighted

ReLAION-natural

Naturalness scores for every image in ReLAION-2B, plus CLIP embeddings for the natural photos.

What it is

Web-scale image datasets mix photographs with screenshots, product shots, memes and text. ReLAION-natural scores all 2.1 billion images in ReLAION-2B for how natural they are, so you can pull out the roughly 500 million natural photographs.

What I did

I trained the naturalness classifier on CLIP embeddings, with about 26,000 images labelled through active learning, and built the GPU pipelines that scored the full dataset and extracted CLIP embeddings for the natural photos.

What came out of it

It's the image pool behind the CCN 2025 paper on sampling natural images and the natural photographs in LAION-fMRI. Together with its embeddings, it has been downloaded about 33,000 times on Hugging Face, currently around 3,800 a month.

View datasetCCN 2025 paper

LAION-fMRI

A deeply sampled fMRI dataset of natural images.

What it is

A 7T fMRI dataset in which five people each spent about 60 hours in the scanner looking at 25,052 natural images, many of them shown several times. It is built to cover far more of the visual world than earlier datasets: 2.5× the effective dimensionality of NSD and THINGS.

What I did

I designed the stimulus set and built the tools around the data: the Python package, the website, the 3D brain explorer and the AWS data access.

What came out of it

It launched at VSS 2026 and is freely available for research; the dataset paper is in preparation. It is also the dataset behind re:vision.

Explore projectBrain explorer

re:vision

A community challenge testing whether findings in visual neuroscience replicate on LAION-fMRI.

What it is

Teams pick a published finding from visual neuroscience and test whether it replicates on LAION-fMRI's much broader set of images, and whether it generalizes. Every valid replication goes into a consortium paper, whatever the result.

What I did

I am one of the coordinators. I made the data splits, ran the benchmark replication of 118 vision models for the consortium paper, and built the website and data services.

What came out of it

It launched at VSS in May 2026, with prizes for the best replication and the best generalization. Reports are due in February 2027, and the synthesis paper follows later that year.

Explore project

Conwell replication

A replication on LAION-fMRI of a large study on which design choices make vision models brain-like.

What it is

Conwell et al. (2024) compared a large survey of vision models against brain responses in NSD to ask which design choices, from architecture to training task and training data, make a model more brain-like. The replication asks whether those controlled comparisons hold on LAION-fMRI's much broader set of images.

What I did

I re-ran the study's own feature-extraction protocol on 118 models and five LAION-fMRI participants, reproduced its controlled comparisons of architecture, training task and training data, and added tests on out-of-distribution images.

What came out of it

It is the benchmark replication for the re:vision consortium paper. Some effects hold, like the advantage of contrastive self-supervised training; others shrink or flip.

View on GitHubOriginal study

thingsvision

An open-source library for extracting and comparing features from vision models.

What it is

Give it a folder of images and a vision model, and it returns the model's representation of every image, with tools to compare those representations to brains and behaviour. It supports models from torchvision, timm, OpenCLIP, Keras and more.

What I did

I am one of its main contributors: I added low-memory feature extraction, support for HDF5 datasets, and the tests that check every integrated model.

What came out of it

It's downloaded about 470 times a month from PyPI and has 179 stars on GitHub.

View library

Say hi

johannes@roth24.de

For research, collaborations, or a good conversation.

Available from January 2027