Soos Talks

Every two weeks, a VIS Lab member presents their research, recent results, or a topic they're exploring over sandwiches in an informal setting.

Why “Soos”? In Dutch, a soos is a social club or meeting place, an informal shortening of sociëteit (social society). Think of a youth club (jeugdsoos) or community hall where people gather for conversation, events, and shared activities. That spirit carries over here. Soos is the lab’s regular gathering point: a place to share work, hear what others are building, and stay connected. I organize the series, including speakers, sandwiches, and logistics, and have hosted over 20 talks during the past year across computer vision, multimodal learning, generative models, and more.

Below is a rolling archive of talks from the series — each card shows the title slide from that session.

Model Correction with Machine Unlearning

Christoforos Spartalis

Jun 02, 2026

Semantic Steerability for Adaptive-Length Image Tokenization

Danilo de Goede

Jun 02, 2026

Hyperbolic Learning for Robust Medical Image Segmentation

Swasti Mishra

May 19, 2026

Abstract Concept Understanding in the Era of Foundation Models

Gowreesh Mago

May 19, 2026

Rewards from Vision-Language Models for Robot Learning

Christian Gumbsch

Apr 21, 2026

HyperHELM: Hierarchy-aligned Language Modeling for mRNA Sequences

Max van Spengler

Apr 21, 2026

Context-Weighted Discrete Flow Matching

Alex Cherniavskii

Apr 07, 2026

Learning to Detect… to Generate Visual Content

Dimitris Karageorgiou

Apr 07, 2026

RFDM: Residual Flow Diffusion Model for Efficient Causal Video Editing

Mohammadreza Salehi

Mar 10, 2026

Physics Grounded World Models for VLAs

Leonard Barcellona

Mar 10, 2026

Elastic ViTs from Pretrained Models without Retraining

Walter Simoncini

Feb 24, 2026

From Deep Learning in Radiotherapy Treatment Planning Towards an Agentic Radiotherapist

Stefanos Achlatis

Feb 10, 2026

In-Context Imitation Learning with E2E-TTT

Riccardo Valperga

Jan 27, 2026

Best Practices in AI Research

Samuele Papa

Dec 09, 2025

Models vs. Brains in Video Understanding

Christina Sartzetaki

Dec 09, 2025

RegionReasoner: Region-Grounded Multi-Round Visual Reasoning

Wenfang Sun

Nov 18, 2025

Instructions: Learning Concepts with Data and Background Knowledge

Emile van Krieken

Oct 07, 2025

NeoBabel: A Multilingual Open Foundation Model for Visual Generation

Dheeraj Varghese

Oct 07, 2025