Spoke 7: Edge and Exascale AI

In the series of "Fair Spoke Workshops 2024", this workshop will focus about Spoke 7 with experts and distinguished guests.

The vast majority of modern Artificial Intelligence is computationally demanding at every stage, from algorithm design and development to training and deployment. Designing new algorithms often involves heuristics and lengthy trial-and-error processes; training requires large-scale data processing, and ensuring strong final performance is closely tied to extensive research and optimization for the best hyperparameters.

This applies to all types of hardware supporting Artificial Intelligence, from cloud and high-performance computing with large-scale data storage to edge devices with limited computational and memory resources. These two computing framework extremes - the infinitely small on the edge and the infinitely large on HPC - are the two emerging computing paradigms for the coming decades. Designing and studying new AI algorithms capable of inherently leveraging the properties of edge and exascale hardware is an open and crucial challenge. There is a need for a new generation of sparse, robust, adaptive, and accurate algorithms for optimization and learning that can support intelligent systems and autonomous agents with varying degrees of supervision. Applications range from industrial robotics to banking, mobility and defense, energy management, and healthcare.

This is the program of the day:

• 9.00 | Welcome and Introduction
  • Barbara Caputo, Politecnico di Torino
• 9.30 | Tiny Machine Learning
  • Speaker: Luciano Prono, Politecnico di Torino
  • Contributions:
    • Regularization for Binary DNNs, T. Bianchi
    • Pruning at Initialization, M. Ciccone, T. Tommasi
    • The MAM Neuron for Prunable DNNs, L. Prono, F. Pareschi, G. Setti
    • Egocentric Videos for Robotics, G. Averta
    • Physics-informed DNNs for Quadrotor Dynamical Modeling, A. Rizzo
• 10.15 | Q&A
• 10.30 | Spotlights Coffee Break
  • Egocentric Vision on the Edge, G. Averta
  • Physics-informed DNNs for Quadrotor Dynamical Modeling, A. Rizzo
  • AI Safety and Reliability, A. Ruospo, L. Sterpone, M. Sonza Reorda
  • Accelerating Heterogeneous Federated Learning with Closed-form Classifiers, M. Ciccone
  • Open-Set Learning, T. Tommasi
• 10.45 | Large Scale Computing
  • Speaker: Francesca Pistilli, Politecnico di Torino
  • Contributions:
    • Learning in High Dimensions, A. Braunstein, A. Pagnani
    • Graph-Informed Neural Networks for Large Scale Applications, S. Pieraccini
    • Federated Learning, M. Ciccone
    • Efficient 3D Semantic Novelty Detection, T. Tommasi
    • GeoAI, A. Lingua, F. Matrone, M. Marano, M.M. Rosso
    • Parallel Learning for Geolocalization, C. Masone, R. Zaccone
• 11.30 | Q&A
• 11.45 | Spotlights Coffee Break
  • Efficient 3D Semantic Novelty Detection, T. Tommasi
  • Semantic Scene Understanding with Urban and Cultural Heritage Point Clouds, F. Matrone
  • Worldwide Visual Place Recognition via Parallel Training, C. Masone, R. Zaccone
  • AI-based Energy Decision Support Systems in Buildings, A. Capozzoli
  • Exploring and Mitigating Bias in Speech Models, E. Baralis
• 12.00 | Round Table on Edge and Exascale AI
• 13.00 | Conclusions

Registration required for online attendance: click here.