Tübingen AI Center Celebrates 38 NeurIPS Contributions at Preview Event with Local AI Community

Showcasing Innovation and Excellence in AI Research

The upcoming 37th NeurIPS conference, the premier global event in machine learning set to take place in New Orleans, USA,marks a significant achievement for the Tübingen AI Center. A total of 38 contributions from its faculty have been accepted following a stringent peer-review process, showcasing the depth and breadth of the institute's expertise.

The Center’s oral and visual presentations span the entire spectrum of AI research. The topics covered are extensive and diverse, including neuro AI, large language models, compositional generalization, computer vision, reinforcement learning, adversarial neural networks, text-to-image diffusion, and causality, among others. For example, in "Propile: Probing privacy leakage in large language models", Siwon Kim and colleagues introduce a novel probing tool designed to empower data subjects to assess the risk that their personally identifiable information could be leaked by an LLM. In “Compositional Generalization from First Principle”, Thaddäus Wiedemer and his co-authors analyse under what conditions we can expect machine learning models to generalise compositionally, a hallmark of human perception. And in “Object-Centric Learning for Real-World Videos by Predicting Temporal Feature Similarities”, Andrii Zadaianchuk and colleagues scale object-centric learning to unconstrained natural videos for the first time. A list of all our talks and posters is available in the Downloads section below.

Workshop on Neural Scaling Laws

As a further contribution, the Tübingen AI Center organizes an important side-event of the conference: Together with Canada Excellence Research Chairs, we will hold the “6th Workshop on Neural Scaling Laws: Scaling, Alignment & Open-Source AI”. It takes place on Friday, December 15, in the Marriott Warehouse Arts District Hotel, New Orleans (please register here). The workshop is sponsored by CIFAR, the Université de Montreal, cerebras and together.ai. The series was initiated in 2021 by CERC AAI Lab with the aim of providing a forum for discussing the most recent advances in large-scale pretrained models.

The workshop will feature the latest developments in large-scale pretrained models, particularly the empirical scaling laws related to performance growth with increased computing power, model size, and pretraining data. It will also address the balance between enhanced AI capabilities and safety/alignment with human values, considering various evaluation metrics. The discussions will relate to transfer, continual, and meta-learning, as well as out-of-distribution generalization, robustness, and invariant/causal predictive modeling.

NeurIPS Preview Events in Tübingen and at several ELLIS Units across Europe

On December 5, the Tübingen AI Center, the ELLIS Institute Tübingen, the ELLIS Unit Tübingen and the Cluster of Excellence “Machine Learning - New Perspectives for Science” at Tübingen University celebrated the conference contributions at a preview event, inviting all AI researchers from Tübingen to view the accepted posters in the building of the new ELLIS Institute. Here the conference participants from Tübingen had a chance to discuss their submissions with their peers one last time before the conference. This gathering took place in parallel to similar events by ten ELLIS Units across Europe.

As AI technology continues to grow and change, the Tübingen AI Center maintains its position as an important contributor to the field. The Center's participation in NeurIPS serves as a testament to its engagement in the broader AI research community, striving to balance innovation with responsibility in technology development.

Get the Tübingen AI Center  News feed