Publications

Which Transformer to Favor: A Comparative Analysis of Efficiency in Vision Transformers

A comprehensive benchmark and analysis of more than 45 transformer models for image classification to evaluate their efficiency, considering various performance metrics. We find the optimal architectures to use and uncover that model-scaling is more efficient than image scaling.

Tobias Christian Nauen, Sebastian Palacio, Federico Raue, Andreas Dengel

Which Transformer to Favor: A Comparative Analysis of Efficiency in Vision Transformers

TaylorShift: Shifting the Complexity of Self-Attention from Squared to Linear (and Back) using Taylor-Softmax

This paper introduces TaylorShift, a novel reformulation of the attention mechanism using Taylor softmax that enables computing full token-to-token interactions in linear time. We analytically and empirically determine the crossover points where employing TaylorShift becomes more efficient than traditional attention. TaylorShift outperforms the traditional transformer architecture in 4 out of 5 tasks.

Tobias Christian Nauen, Sebastian Palacio, Andreas Dengel

Zoomed In, Diffused Out: Towards Local Degradation-Aware Multi-Diffusion for Extreme Image Super-Resolution

We extend pretrained super-resolution models to larger images by using local-aware prompts.

Brian B. Moser, Stanislav Frolov, Tobias Christian Nauen, Federico Raue, Andreas Dengel

Zoomed In, Diffused Out: Towards Local Degradation-Aware Multi-Diffusion for Extreme Image Super-Resolution

Just Leaf It: Accelerating Diffusion Classifiers with Hierarchical Class Pruning

We speed up diffusion classifiers by utilizing a label hierarchy and pruning unrelated paths.

Arundhati S Shanbhag, Brian Bernhard Moser, Tobias Christian Nauen, Stanislav Frolov, Federico Raue, Andreas Dengel

Just Leaf It: Accelerating Diffusion Classifiers with Hierarchical Class Pruning

Distill the Best, Ignore the Rest: Improving Dataset Distillation with Loss-Value-Based Pruning

We improve dataset distillation by distilling only a representative coreset.

Brian Bernhard Moser, Federico Raue, Tobias Christian Nauen, Stanislav Frolov, Andreas Dengel

Distill the Best, Ignore the Rest: Improving Dataset Distillation with Loss-Value-Based Pruning

A Low-Resolution Image is Worth 1x1 Words: Enabling Fine Image Super-Resolution with Transformers and TaylorShift

We utilize the TaylorShift attention mechanism for global pixel-wise-attention in image super-resolution.

Sanath Budakegowdanadoddi Nagaraju, Brian Bernhard Moser, Tobias Christian Nauen, Stanislav Frolov, Federico Raue, Andreas Dengel

Stochastic Control with Signatures

This paper proposes a new method to parameterize open loop controls in stochastic optimal control problems using path signatures. We show that these controls are dense in the space of all admissible controls and establish conditions for stability of the controlled dynamics and target functional.

Peter Bank, Christian Bayer, Paul Peter Hager, Sebastian Riedel, Tobias Christian Nauen

Stochastic Optimal Control using Signatures

We consider a stochastic control problem and try to solve it using the signature method.

Tobias Christian Nauen, Sebastian Riedel

Stochastic Optimal Control using Signatures

Explaining Graph Neural Networks

We extend and test KEdge, an interpretable-by-design approach for graph neural networks, and compare it to gradient-based attribution techniques.

Tobias Christian Nauen, Thorben Funke, Avishek Anand

Explaining Graph Neural Networks