Frederik Träuble

Co-Founder at calliora

[Frederik Träuble @f_traeuble]

Check out this exciting work, led by the exceptional @FrederikeLubeck if you are at ICML 2025!

[Frederike Lübeck @FrederikeLubeck]

Clinical notes are messy, inconsistent, and unstructured—yet they hold some of the most valuable signals in real-world clinical practice.

Join us today at ICML at the Foundation Models for Structured Data workshop to see how we can make sense of these notes!
📍 West Ballroom D

[Frederik Träuble @f_traeuble]

Easily one of the biggest and most fascinating projects I’ve ever worked on—huge thanks to @schwabpa and the whole team for having me and the opportunity to collaborate during my time @GSK. Check out all the details in this summary!

Preprint: https://t.co/6DAJW2RWDu

[Patrick Schwab @schwabpa]

Understanding human biology across scales - from molecules to cells to entire organisms - remains one of biomedicine's greatest challenges in the fight against disease.

Today, we are announcing Phenformer - a multi-scale genetic language model that learns to read and interpret human genomes by connecting DNA, cell and tissue context, molecules and clinical outcomes.

Phenformer is a generative model of molecular mechanisms that enables researchers to unravel the mysteries underlying disease, and could thereby accelerate the development of precise future therapeutics.

[Vedant Shah @veds_12]

Unlearning via Sparse Representations

Work with @f_traeuble, Ashish Malik, @hugo_larochelle, @MichaelMozer1, @prfsanjeevarora, Yoshua Bengio & @anirudhg9119

https://t.co/6edmvYc8LB

TLDR: We propose a new approach enabling nearly compute-free class unlearning. 🧵

[Cian Eastwood @CianEastwood]

Introducing our #ICLR2023 paper:
DCI-ES: An Extended Disentanglement Framework with Connections to Identifiability🚀

We propose a new notion of disentanglement based on the functional capacity required to use a representation

https://t.co/9c3WGeeajS
https://t.co/9MYWmbTJMY
1/12

[Patrick Schwab @schwabpa]

Creating a map of gene interactions is a fundamental step in drug discovery that generates ideas on what mechanisms may be targeted by future medicines

Today, we announce the CausalBench challenge at https://t.co/yyLDX9yR1b and invite you to contribute to this important problem!

[Anirudh Goyal @anirudhg9119]

Discrete Key-Value Bottleneck (Updated)

Compresses the information of a pre-trained model in learnable "key-value" codebook such that knowledge can be quickly adapted in a continual learning fashion.

https://t.co/RdgOaBStDJ

[Armin Kekić @armin_kekic]

Many countries employed an age-ranked vaccine allocation strategy to combat COVID-19. How effective was this strategy at preventing infections and severe cases? We study this and other questions using simulation-assisted causal modelling. 🧵 1/

preprint: https://t.co/PJPvM53lua

[Nasim Rahaman @nasim_rahaman]

🥁We raise you Neural Attentive Circuits, a general-purpose modular neural architecture.

With Martin Weiss (co-lead), @FrancescoLocat8, @chrisjpal, Yoshua Bengio, @bschoelkopf, Erran Li & Nicolas Ballas. @Mila_Quebec @MPI_IS @MetaAI @AmazonScience

https://t.co/yJygEtSfWy

🧶

[Causal Representation Learning Workshop @ UAI'22 @crl_uai]

Only 1 week left until the 1st Workshop on Causal Representation Learning at @UncertaintyInAI

Lists of accepted papers & reviewers, additional information on how to attend, and a detailed schedule incl. speakers are now available on the workshop website: https://t.co/G8UJGdzTfe

[Frederik Träuble @f_traeuble]

Most importantly, this is joint work and wouldn’t have been possible without fantastic co-authors @anirudhg9119 @nasim_rahaman @mc_mozer Kenji Kawaguchi, Yoshua Bengio, @bschoelkopf.

@MPI_IS @Mila_Quebec

Any feedback is much appreciated.🙂

6/6

[Frederik Träuble @f_traeuble]

“Discrete Key-Value Bottlenecks”

Amortizing information via a discrete bottleneck such that the knowledge is localized and results in flexible adaptation to distribution shifts such as non-stationary or imbalanced data streams.

https://t.co/o7jSAIpiHx
1/6

[Frederik Träuble @f_traeuble]

By freezing all model parameters except for the value codes, we can keep learning under various distribution shifts. This is enabled via localized, input-dependent model updates, which don't affect the prediction from (key, value) pairs retrieved from unalike train samples.

5/6

[Cian Eastwood @CianEastwood]

Look forward to presenting our work! 🚀

We connect the DCI disentanglement scores to identifiability, and propose a new complementary notion of disentanglement based on the *functional capacity required to use a representation.*

🔗https://t.co/RyVRcjUUil
🧵Short thread below

[Frederik Träuble @f_traeuble]

Check out our latest work if you have ever struggled with learning RL agents that can solve dexterous object-manipulation tasks in multi-object robotics settings! 🤖

https://t.co/HKkof5I2p7

Joint work with @mambelli_davide, Stefan Bauer, @bschoelkopf and @FrancescoLocat8

[Francesco Locatello @FrancescoLocat8]

"Compositional Multi-Object Reinforcement Learning with Linear Relation Networks": new module with linear cost and an object-centric compositional bias for training RL policies that generalize zero shot to arbitrary number of objects in robotics. [1/2]

https://t.co/gcDkq1b0df

[Francesco Locatello @FrancescoLocat8]

“Visual Representation Learning Does Not Generalize Strongly Within the Same Domain”: regardless of architecture and training signal, deep nets struggle to generalize strongly to existing factors of variation in the training data.

https://t.co/9Xz2VvaO6U

[Frederik Träuble @f_traeuble]

Happy to share that our work on "The Role of Pretrained Representations for the OOD Generalization of RL Agents" was accepted to #iclr2022! 🎉

[Andrea Dittadi @andrea_dittadi]

Happy to announce our large-scale study on representation learning and generalization in reinforcement learning! https://t.co/h9fSpw54Iw

How do properties of pre-trained representation backbones affect the robustness of downstream RL policies in simulation and real world?

1/5

[Tom Goldstein @tomgoldsteincs]

My recent talk at the NSF town hall focused on the history of the AI winters, how the ML community became "anti-science," and whether the rejection of science will cause a winter for ML theory. I'll summarize these issues below...🧵