Explore the cutting-edge research programs and projects led by the ELLIS Italy network. Discover the innovative work being done in artificial intelligence, machine learning, and data science across our member institutions.
This program aims to establish geometric principles such as structure, symmetry, and invariance as a unifying foundation for deep learning. Its research pillars include the mathematical formalization of neural architectures, geometric inductive biases, unification of CNNs, GNNs, and Transformers, principled model design, and theory-driven development of next-generation AI systems.
This program aims to transform biomedicine and healthhcare through cutting-edge AI and machine learning research. Its research pillars include AI-driven disease modeling and diagnosis, data-centric and trustworthy health AI, integration of multimodal biomedical data, translational collaboration between AI and clinical experts, and training of the next generation of AI-health researchers.
This program aims to ensure that machine learning systems generate broad societal benefits and earn public trust. Its research pillars include transparency and interpretability of algorithmic decisions, fairness and accountability, legal and technical certification frameworks, robustness in real-world deployment, and human oversight in AI systems.
This program aims to advance interactive models of learning by integrating causal reasoning into adaptive intelligent systems. Its research pillars include causal modeling for interventional learning, bridging observational and experimental data, interactive decision-making, robustness in high-stakes environments, and principled foundations of reliable learning-based systems.
This program aims to integrate machine learning with computer graphics, geometry processing, and 3D computer vision. Its research pillars include geometric representations for learning, differentiable rendering and simulation, 3D scene understanding, learning-enhanced graphics pipelines, and cross-fertilization between vision, geometry, and deep learning.
This program aims to foster scientific exchange and collaboration in machine learning and computer vision. Its research pillars include visual representation learning, large-scale vision models, multimodal perception, theoretical foundations of visual understanding, and real-world deployment of computer vision systems.
This program aims to model and understand the Earth system by combining machine learning with physical and process-based knowledge. Its research pillars include spatio-temporal modeling of climate dynamics, extreme event detection and attribution, hybrid physics-informed learning, generative Earth system modeling, data–model fusion, and interpretable climate forecasting.
This program aims to accelerate molecular discovery through the integration of machine learning and AI with molecular sciences. Its research pillars include generative models for molecule design, property prediction and optimization, AI-driven simulation and screening, integration of chemical and biological data, and collaborative frameworks bridging AI and domain expertise.
This program aims to advance the theory and practice of multimodal learning systems across data modalities and application domains. Its research pillars include multimodal representation learning, foundation models integrating vision, language and other modalities, cross-modal alignment and reasoning, scalable training methodologies, and real-world multimodal applications.
This program aims to advance the science of natural intelligence by studying how brains learn and generalize across tasks and environments. Its research pillars include lifelong and continual learning, agent-centric intelligence, adaptive computation, inductive biases and deep semantics, social reasoning, and biologically inspired models of cognition.
The members of the ELLIS program ‘Learning for Graphics and Vision’ see a big opportunity to integrate classical computer graphics, geometry processing and 3D vision principles more tightly within deep learning frameworks. The mission of this program is to connect researchers in the fields of machine learning, computer graphics, 3D computer vision, and geometry processing to discuss and investigate how machine learning can benefit from computer graphics, geometry processing and 3D vision, as well as to investigate the explosion of new possibilities that machine learning enables for graphics and vision.
This program aims to develop novel machine learning methods inspired by quantum and statistical physics. Its research pillars include quantum-enhanced learning algorithms, energy-efficient physical implementations, physics-informed model design, applications of machine learning to quantum information processing, and exploration of new computational paradigms.
This program aims to close the gap between simulated and real-world robotic systems by advancing principled approaches to robot learning and control. Its research pillars include robot motion and action generation, learning-based interaction, sensorimotor adaptation, sim-to-real transfer, and machine learning methods to improve robustness and real-world performance.
This program aims to establish principled foundations for robust machine learning in complex and safety-critical environments. Its research pillars include robustness to distribution shifts and adversarial perturbations, formal verification and quantification of reliability, uncertainty estimation, trustworthy deployment in real-world systems, and applications in healthh, environmental sciences, autonomous systems, and industrial control.
This program aims to advance semantic, symbolic, and interpretable machine learning methods that operate at the level of human-understandable abstractions. Its research pillars include multi-relational learning, temporal and structured knowledge graphs, extraction of statistical and logical regularities from data, embedding and graph-based representations, neuro-symbolic and inductive logic programming approaches, and cross-fertilization with related areas such as natural language processing, vision, and geometric deep learning.
This program aims to strengthen the theoretical foundations of modern machine learning systems by bridging the gap between empirical success and formal understanding. Its research pillars include mathematical analysis of contemporary architectures, algorithmic principles of large-scale learning systems, computational complexity and optimization theory, generalization and stability guarantees, and the development of a unified theoretical framework connecting diverse strands of theoretical machine learning.
