A5083981921- Advanced Graph Neural Networks
- Topic Modeling
- Machine Learning in Materials Science
- Natural Language Processing Techniques
- Explainable Artificial Intelligence (XAI)
- Machine Learning in Healthcare
- Aerospace and Aviation Technology
- Human-Automation Interaction and Safety
- Advanced Text Analysis Techniques
- Multimodal Machine Learning Applications
- Infrared Target Detection Methodologies
- X-ray Diffraction in Crystallography
- Glaucoma and retinal disorders
- Nuclear Physics and Applications
- Spatial Cognition and Navigation
- Catalytic Processes in Materials Science
- Computational Drug Discovery Methods
- Visual perception and processing mechanisms
- Modular Robots and Swarm Intelligence
- Retinal Development and Disorders
- Graph Theory and Algorithms
- Complex Network Analysis Techniques
- Cloud Computing and Resource Management
- Robot Manipulation and Learning
École de Technologie Supérieure
2023
CentraleSupélec
2021-2023
Bouygues (France)
2021-2023
Mila - Quebec Artificial Intelligence Institute
2023
Université Paris-Saclay
2021-2022
École Centrale d'Électronique
2021
Naver (South Korea)
2021
University of Sheffield
2019
Graph Neural Networks achieve state-of-the-art performance on a plethora of graph classification tasks, especially due to pooling operators, which aggregate learned node embeddings hierarchically into final representation. However, they are not only questioned by recent work showing par with random pooling, but also ignore completely higher-order connectivity patterns. To tackle this issue, we propose HoscPool, clustering-based operator that captures information hierarchically, leading...
Recent advances in computational modelling of atomic systems, spanning molecules, proteins, and materials, represent them as geometric graphs with atoms embedded nodes 3D Euclidean space. In these graphs, the attributes transform according to inherent physical symmetries including rotations translations space, well node permutations. recent years, Geometric Graph Neural Networks have emerged preferred machine learning architecture powering applications ranging from protein structure...
Alexandre Duval, Thomas Lamson, Gaël de Léséleuc Kérouara, Matthias Gallé. Proceedings of the 16th Conference European Chapter Association for Computational Linguistics: System Demonstrations. 2021.
Applications of machine learning techniques for materials modeling typically involve functions known to be equivariant or invariant specific symmetries. While graph neural networks (GNNs) have proven successful in such tasks, they enforce symmetries via the model architecture, which often reduces their expressivity, scalability and comprehensibility. In this paper, we introduce (1) a flexible framework relying on stochastic frame-averaging (SFA) make any E(3)-equivariant through data...
Rapid advancements in machine learning (ML) are transforming materials science by significantly speeding up material property calculations. However, the proliferation of ML approaches has made it challenging for scientists to keep with most promising techniques. This paper presents an empirical study on Geometric Graph Neural Networks 3D atomic systems, focusing impact different (1) canonicalization methods, (2) graph creation strategies, and (3) auxiliary tasks, performance, scalability...
The drone industry is diversifying and the number of pilots increasing rapidly. In this context, flight schools need adapted tools to train pilots, most importantly with regard their own awareness physiological cognitive limits. civil military aviation, can on realistic simulators tune reaction reflexes, but also gather data piloting behavior states, helping improve performance. As opposed cockpit scenarios, teleoperation conducted outdoors in field, only limited potential from desktop...
Mitigating the climate crisis requires a rapid transition towards lower-carbon energy. Catalyst materials play crucial role in electrochemical reactions involved numerous industrial processes key to this transition, such as renewable energy storage and electrofuel synthesis. To reduce spent on activities, we must quickly discover more efficient catalysts drive reactions. Machine learning (ML) holds potential efficiently model properties from large amounts of data, accelerating...
The use of machine learning for material property prediction and discovery has traditionally centered on graph neural networks that incorporate the geometric configuration all atoms. However, in practice not this information may be readily available, e.g.~when evaluating potentially unknown binding adsorbates to catalyst. In paper, we investigate whether it is possible predict a system's relaxed energy OC20 dataset while ignoring relative position adsorbate with respect electro-catalyst. We...
Accelerating material discovery holds the potential to greatly help mitigate climate crisis. Discovering new solid-state materials such as electrocatalysts, super-ionic conductors or photovoltaic can have a crucial impact, for instance, in improving efficiency of renewable energy production and storage. In this paper, we introduce Crystal-GFN, generative model crystal structures that sequentially samples structural properties crystalline materials, namely space group, composition lattice...
The drone industry is diversifying and the number of pilots increases rapidly. In this context, flight schools need adapted tools to train pilots, most importantly with regard their own awareness physiological cognitive limits. civil military aviation, can themselves on realistic simulators tune reaction reflexes, but also gather data piloting behavior states. It helps them improve performances. Opposed cockpit scenarios, teleoperation conducted outdoor in field, thus only limited potential...
Graph Neural Networks (GNNs) achieve significant performance for various learning tasks on geometric data due to the incorporation of graph structure into node representations, which renders their comprehension challenging. In this paper, we first propose a unified framework satisfied by most existing GNN explainers. Then, introduce GraphSVX, post hoc local model-agnostic explanation method specifically designed GNNs. GraphSVX is decomposition technique that captures "fair" contribution each...
It is standard procedure these days to solve Information Extraction task by fine-tuning large pre-trained language models. This not the case for generation task, which relies on a variety of techniques controlled generation. In this paper, we describe system that fine-tunes natural model problem solving Writer's Block. The changes conditioning also include right context in addition left context, as well an optional list entities, size, genre and summary paragraph human author wishes...