Every year, MDSI awards three outstanding papers published in the previous year by groups of MDSI Core Members. Again, first place was awarded €1,000, the runner-up received €500, and third place received an honorable mention. All three winners received a certificate from MDSI.

First Place

The paper on "DiffRF: Rendering-Guided 3D Radiance Field Diffusion", awarded with the fist prize, describes a method that enables the generation of 3D models such as different types of chairs and tables from simple natural language prompts.
First author Norman Müller and his colleagues from the Matthias Nießner Group at TUM and the Meta Reality Labs Zurich present the rendering of models from any selected viewpoint, hence providing new possibilities over image models that restrict synthesis to a single, fixed view. Their approach is based on the "denoising-diffusion" methodology and involves the training of a model to gradually denoise a diffused input enabling the synthesis of high-fidelity images from Gaussian with noise. 
The publication was recognized as a "Highlight" by the Conference on Computer Vision and Pattern Recognition and became the most popular ML arXiv paper when it appeared.

Runner up 

In her publication "Graph-in-Graph (GiG): Learning interpretable latent graphs in non-Euclidean domain for biological and healthcare applications", awarded with the second prize, Kamilia Zaripova and colleagues from the Nassir Navab Group introduce a novel graph neural network architecture. The model termed Graph-in-Graph (GiG), was designed to exploit the graph representation of input data samples and their latent relations, enabling knowledge discovery.
GiG is the first method to leverage graph data as input directly to obtain the population/relationship graph, allowing for modelling complex dependencies in a novel and theoretically compelling manner. The interdisciplinary application of GiG to biological and healthcare data demonstrates its utility in critical areas affecting human health, bridging advanced machine learning techniques with practical healthcare applications. Additionally, GiG addresses the challenge of model trustworthiness in healthcare by making latent structures explicit.

Honorable Mention

Arthur Kosmala and his colleagues from the Stephan Günnemann Group published their interdisciplinary work on "Ewald-based Long-Range Message Passing for Molecular Graphs" and provided a stepstone towards uniting accuracy and efficiency of neural property prediction in the wide spectrum of systems impacted by long-range interactions.
In its prominent role as a "computational microscope", molecular modelling has vitally advanced our capacity to predict accessible properties of molecules and materials from small-scale insights into their atomistic behaviour. The principled theory behind Kosmalas method, along with the significant improvements achieved by its practical implementation, reflects an insightful combination of physical domain knowledge applied with cutting-edge machine learning expertise. 

Links:

1. First place: Norman Müller et al., "DiffRF: Rendering-Guided 3D Radiance Field Diffusion", CVPR 2023. 

2. Runner up: Kamilia Zaripova et al. "Graph-in-Graph (GiG): Learning interpretable latent graphs in non-Euclidean domain for biological and healthcare applications", MedAI 2023.

3. Honorable Mention: Arthur Kosmala et al. "Ewald-based Long-Range Message Passing for Molecular Graphs", PMLR 202:17544-17563, 2023.