A5084534227- Advanced Graph Neural Networks
- Distributed and Parallel Computing Systems
- Scientific Computing and Data Management
- Neural Networks and Applications
- Medical Image Segmentation Techniques
- Text and Document Classification Technologies
- 3D Surveying and Cultural Heritage
- Geological Modeling and Analysis
- Model Reduction and Neural Networks
- Image and Object Detection Techniques
- Neural dynamics and brain function
- Image Processing and 3D Reconstruction
- Graph Theory and Algorithms
- Neural Networks and Reservoir Computing
- Rough Sets and Fuzzy Logic
- Machine Learning and Data Classification
- Machine Learning in Materials Science
- Cell Image Analysis Techniques
- 3D Shape Modeling and Analysis
- Bayesian Modeling and Causal Inference
- Computational Drug Discovery Methods
University of Milan
2023-2025
Abstract Recent advances in machine learning have highlighted the importance of using group equivariant non-expansive operators for building neural networks a more transparent and interpretable way. An operator is called with respect to if action commutes operator. Group can be seen as multi-level components that joined connected order form by applying operations chaining, convex combination direct product. In this paper we prove each linear G -equivariant (GENEO) produced weighted summation...
Nowadays there is a big spotlight cast on the development of techniques explainable machine learning. Here we introduce new computational paradigm based Group Equivariant Non-Expansive Operators, that can be regarded as product rising mathematical theory information-processing observers. This approach, adjusted to different situations, may have many advantages over other common tools, like Neural Networks, such as: knowledge injection and information engineering, selection relevant features,...
The theory of Group Equivariant Non-Expansive Operators (GENEOs) was initially developed in Topological Data Analysis for the geometric approximation data observers, including their invariances and symmetries. This paper departs from that line research explores use GENEOs distinguishing $r$-regular graphs up to isomorphisms. In doing so, we aim test capabilities flexibility these operators. Our experiments show offer a good compromise between efficiency computational cost comparing graphs,...
The study of $G$-equivariant operators is great interest to explain and understand the architecture neural networks. In this paper we show that each linear operator can be produced by a suitable permutant measure, provided group $G$ transitively acts on finite signal domain $X$. This result makes available new method build in setting.
Research in 3D semantic segmentation has been increasing performance metrics, like the IoU, by scaling model complexity and computational resources, leaving behind researchers practitioners that (1) cannot access necessary resources (2) do need transparency on decision mechanisms. In this paper, we propose SCENE-Net, a low-resource white-box for point cloud segmentation. SCENE-Net identifies signature shapes via group equivariant non-expansive operators (GENEOs), providing intrinsic...