Frederik Träuble

I am a PhD student at the Max Planck Institute for Intelligent Systems, advised by Bernhard Schölkopf. I was fortunate to spend time at Mila working in the group of Yoshua Bengio as well as AWS AI and GSK AI during my PhD.

I am mainly interested in anything related to scaling and (out-of-distribution) generalization of deep neural networks and worked across diverse applications such as computer vision, robotics and most recently population-scale genomics data.

Before starting my PhD, I studied Physics and Scientific Computing in Munich and Cambridge.

LinkedIn | GitHub | Twitter | GoogleScholar

News

Apr: 2023: Our paper "Discrete Key-Value Bottleneck" was accepted at ICML 2023
Jul 2022: I started a summer internship with the Biomedical AI team at GSK.ai (GlaxoSmithKline) training large models on huge datasets.
Apr 2022: Two papers at ICLR 2022: "The Role of Pretrained Representations for the OOD Generalization of Reinforcement Learning Agents" and "Visual Representation Learning Does Not Generalize Strongly Within the Same Domain".
Oct 2021: I am visiting Mila in Montreal as Visiting Researcher working with Anirudh Goyal and Yoshua Bengio.
Sep 2021: Our paper "Backward-Compatible Prediction Updates: A Probabilistic Approach" was accepted at NeurIPS 2021.
Jul 2021: Our paper "On Disentangled Representations Learned From Correlated Data" was accepted for a long oral presentation (top 3%) at ICML 2021. 🥳
May 2021: Our two papers "CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning" and "On the Transfer of Disentangled Representations in Realistic Settings" have been accepted at ICLR 2021. 🥳
Jan 2021: Our paper "An improved equation of state for air plasma simulations" got featured as Editor's Pick in Physics of Fluids. 🥳
Nov 2020: Our work on CausalWorld got featured in DigitalTrends: This basic human skill is the next major milestone for A.I. link
Oct 2020: Our work on CausalWorld got featured in IEEE Spectrum: Understanding Causality Is the Next Challenge for Machine Learning. link
Sep 2020: I joined Amazon as an Applied Science Intern.

Selected Publications (see Google Scholar for full list)

Discrete Key-Value Bottleneck

Frederik Träuble, Anirudh Goyal, Nasim Rahaman, Michael Mozer, Kenji Kawaguchi, Yoshua Bengio, Bernhard Schölkopf

ICML 2023

[Paper] [Code]

On Disentangled Representations Learned From Correlated Data

Frederik Träuble, Elliot Creager, Niki Kilbertus, Francesco Locatello, Andrea Dittadi, Anirudh Goyal, Bernhard Schölkopf, Stefan Bauer

ICML 2021. Long oral presentation (top 3%)

[Paper] [Code]

Backward-Compatible Prediction Updates: A Probabilistic Approach

Frederik Träuble*, Julius von Kügelgen, Matthäus Kleindessner, Francesco Locatello, Bernhard Schölkopf, Peter Gehler

NeurIPS 2021

[Paper]

The Role of Pretrained Representations for the OOD Generalization of Reinforcement Learning Agents

Frederik Träuble*, Andrea Dittadi*, Manuel Wüthrich, Felix Widmaier, Peter Gehler, Ole Winther, Francesco Locatello, Olivier Bachem, Bernhard Schölkopf, Stefan Bauer

ICLR 2022

[Paper]

ICML 2021 Workshop on Reinforcement Learning for Real Life
ICML 2021 Workshop on Unsupervised Reinforcement Learning

Compositional Multi-Object Manipulation with Linear Relation Networks

Davide Mambelli*, Frederik Träuble*, Stefan Bauer, Bernhard Schölkopf, Francesco Locatello

ICLR 2022 Workshop on the Elements of Reasoning: Objects, Structure and Causality

[Paper]

CausalWorld: A Robotic Manipulation Benchmark for Causal Structure and Transfer Learning

Ossama Ahmed*, Frederik Träuble*, Anirudh Goyal, Alexander Neitz, Yoshua Bengio, Bernhard Schölkopf, Manuel Wüthrich, Stefan Bauer

ICLR 2021

[Paper] [Benchmark]

On the Transfer of Disentangled Representations in Realistic Settings

Andrea Dittadi*, Frederik Träuble*, Francesco Locatello, Manuel Wüthrich, Vaibhav Agrawal, Ole Winther, Stefan Bauer, Bernhard Schölkopf

ICLR 2021

[Paper] [Dataset]

Visual Representation Learning Does Not Generalize Strongly Within the Same Domain

Lukas Schott, Julius von Kügelgen, Frederik Träuble, Peter Gehler, Chris Russell, Matthias Bethge, Bernhard Schölkopf, Francesco Locatello, Wieland Brendel

ICLR 2022

[Paper] [Benchmark]

ICLR 2021 Workshop on Generalization beyond the training distribution in brains and machines

An improved equation of state for air plasma simulations

Frederik Träuble, Stephen Millmore, Nikolaos Nikiforakis

Physics of Fluids 33, 036112. 2021. Featured as Editors’ Pick.

[Paper]

A stepladder approach to tokamak fusion power plant.

Hartmut Zohm, Frederik Träuble, Wolfgang Biel, Emiliano Fable, Richard Kemp, Hanni Lux, Mattia Siccinio, Ronald Wenninger

Nuclear Fusion, Volume 57, Number 8. 2017

[Paper]

Contact

[first][dot]traeuble@tuebingen.mpg.de