MPI-IS researchers with 11 papers at upcoming ICML

As in previous years, MPI-IS researchers will be presenting a large number of papers at the conference, which will take place in Vancouver from 13 to 19 July 2025.

On July 15th, the following posters will be presented:

Generalized Interpolating Discrete Diffusion

Great Models Think Alike and this Undermines AI Oversight

LAION-C: An Out-of-Distribution Benchmark for Web-Scale Vision Models

Position: An Empirically Grounded Identifiability Theory Will Accelerate Self Supervised Learning Research

On July 16th, three posters are in the line-up:

When, Where and Why to Average Weights?

LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws

Partially Observable Reinforcement Learning with Memory Traces

And on July 17th the remaining batch are:

Generative Intervention Models for Causal Perturbation Modeling

Learning Joint Interventional Effects from Single-Variable Interventions in Additive Models

An Interpretable N-gram Perplexity Threat Model for Large Language Model Jailbreaks

Distributed Event-Based Learning via ADMM