A5053890378- Neural Networks and Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Graph Neural Networks
- Neural Networks and Reservoir Computing
- Topic Modeling
- Multimodal Machine Learning Applications
- Music and Audio Processing
- Machine Learning in Materials Science
- Advanced Memory and Neural Computing
- Complex Network Analysis Techniques
- Natural Language Processing Techniques
- Computational Drug Discovery Methods
- Data Stream Mining Techniques
- Anomaly Detection Techniques and Applications
- Tensor decomposition and applications
- Model Reduction and Neural Networks
- Neural dynamics and brain function
- Time Series Analysis and Forecasting
- Generative Adversarial Networks and Image Synthesis
- Graph Theory and Algorithms
- Explainable Artificial Intelligence (XAI)
- Machine Learning and Algorithms
- Music Technology and Sound Studies
- Gene expression and cancer classification
- COVID-19 diagnosis using AI
University of Pisa
2015-2024
University of Rhode Island
2021
Canadian Standards Association
2021
Institute of Electrical and Electronics Engineers
2021
X-Fab (Germany)
2021
University of Padua
2020
IMT School for Advanced Studies Lucca
2006-2009
Scuola Superiore Sant'Anna
2005
Experimental reproducibility and replicability are critical topics in machine learning. Authors have often raised concerns about their lack scientific publications to improve the quality of field. Recently, graph representation learning field has attracted attention a wide research community, which resulted large stream works. As such, several Graph Neural Network models been developed effectively tackle classification. However, experimental procedures rigorousness hardly reproducible....
Learning continually from non-stationary data streams is a long-standing goal and challenging problem in machine learning. Recently, we have witnessed renewed fast-growing interest continual learning, especially within the deep learning community. However, algorithmic solutions are often difficult to re-implement, evaluate port across different settings, where even results on standard benchmarks hard reproduce. In this work, propose Avalanche, an open-source end-to-end library for research...
Soft hands are robotic systems that embed compliant elements in their mechanical design.This enables an effective adaptation with the items and environment, ultimately, increase grasping performance.These come clear advantages terms of ease-to-use robustness if compared classic rigid hands, when operated by a human.However, potential for autonomous is still largely unexplored, due to lack suitable control strategies.To address this issue, letter, we propose approach enable soft autonomously...
Estimation of mortality risk very preterm neonates is carried out in clinical and research settings. We aimed at elaborating a prediction tool using machine learning methods. developed models on cohort 23747 <30 weeks gestational age, or <1501 g birth weight, enrolled the Italian Neonatal Network 2008-2014 (development set), 12 easily collected perinatal variables. used from 2015-2016 (N = 5810) as test set. Among several methods we chose artificial Neural Networks (NN). The resulting...
Online Continual learning is a challenging scenario where the model must learn from non-stationary stream of data each sample seen only once. The main challenge to incrementally while avoiding catastrophic forgetting, namely problem forgetting previously acquired knowledge new data. A popular solution in these use small memory retain old and rehearse them over time. Unfortunately, due limited size, quality will deteriorate In this paper we propose OLCGM, novel replay-based continual strategy...
Recent progress in research on deep graph networks (DGNs) has led to a maturation of the domain learning graphs. Despite growth this field, there are still important challenges that yet unsolved. Specifically, is an urge making DGNs suitable for predictive tasks real-world systems interconnected entities, which evolve over time. With aim fostering dynamic graphs, first, we survey recent advantages both temporal and spatial information, providing comprehensive overview current...
Graph State Space Models (SSMs) have recently been introduced to enhance Neural Networks (GNNs) in modeling long-range interactions. Despite their success, existing methods either compromise on permutation equivariance or limit focus pairwise interactions rather than sequences. Building the connection between Autoregressive Moving Average (ARMA) and SSM, this paper, we introduce GRAMA, a Adaptive method based learnable framework that addresses these limitations. By transforming from static...
The Cloud-Edge continuum enhances application performance by bringing computation closer to data sources. However, it presents considerable challenges in managing resources and determining service placement, as these tasks require navigating diverse, dynamic environments characterised fluctuating network conditions. Addressing calls for tools combining simulation emulation of systems rigorously assess novel resource management strategies. In this paper, we introduce ECLYPSE, a Python-based...
Optimizing chemical properties is a challenging task due to the vastness and complexity of space. Here, we present generative energy-based architecture for implicit property optimization, designed efficiently generate molecules that satisfy target without explicit conditional generation. We use Graph Energy Based Models training approach does not require labels. validated our on well-established benchmarks, showing superior results state-of-the-art methods demonstrating robustness efficiency...
In this work, we introduce HEIMDALL, a grapH-based sEIsMic Detector And Locator specifically designed for microseismic applications. Building on recent progress in deep learning (DL), HEIMDALL employs spatiotemporal graph-neural networks to detect and locate seismic events continuous waveforms. It simultaneously associates provides preliminary locations by leveraging the output probability functions of network over dense, three-dimensional grid (0.1 km spacing). By integrating detection...
In recent decades, geothermal systems have gained increasing importance and attention. They the potential to greatly contribute transition toward green energy establishment of a climate-neutral economy. Enhanced Geothermal Systems (EGS) represent significant advancement in production methodologies. EGS utilize hydraulic stimulation techniques inject extract fluids, thereby enabling harnessing energy, which is crucial for electricity generation.In addition existing natural seismicity, this...
We introduce a novel compositional (recursive) probabilistic model for trees that defines an approximated bottom-up generative process from the leaves to root of tree. The proposed contextual state transitions joint configuration children parent nodes. argue context postulates different assumptions with respect top-down approach, leading representational capabilities. discuss classes applications are best suited approach. In particular, is shown better correlate and co-occurrence...
We introduce the Contextual Graph Markov Model, an approach combining ideas from generative models and neural networks for processing of graph data. It founds on a constructive methodology to build deep architecture comprising layers probabilistic that learn encode structured information in incremental fashion. Context is diffused efficient scalable way across vertexes edges. The resulting encoding used combination with discriminative address structure classification benchmarks.
Pre-trained models are commonly used in Continual Learning to initialize the model before training on stream of non-stationary data. However, pre-training is rarely applied during Learning. We investigate characteristics Pre-Training scenario, where a continually pre-trained incoming data and only later fine-tuned different downstream tasks. introduce an evaluation protocol for which monitors forgetting against Forgetting Control dataset not present continual stream. disentangle impact 3...
Molecule generation is a challenging open problem in cheminformatics. Currently, deep generative approaches addressing the challenge belong to two broad categories, differing how molecules are represented. One approach encodes molecular graphs as strings of text, and learns their corresponding character-based language model. Another, more expressive, operates directly on graph. In this work, we address limitations former: invalid duplicate molecules. To improve validity rates, develop model...