Information Engineering for Sustainable Computing

Introduction

Artificial Intelligence (AI) is developing unprecedentedly nowadays and big data processing necessitates more powerful methods for extracting insights that lead to more precise models for system control and optimization. The huge amount of data available is generated by a plethora of devices geographically distributed such as smartphones, sensors, wearable devices and different types of machines, most of them connected via mobile networks mainly.

However, the accuracy improvements of these AI models is subjected on the usage of very large computational resources that, in turns, drain a considerable amount of energy. For example, training a single AI model for a natural language processing task is equivalent to 284 tonnes of carbon dioxide equivalent, i.e., five times the lifetime emissions of an average car [1].

Therefore, the possibility to process the data directly on the edge than in mega-scale data centres is emerging as an efficient solution to avoid oppressive network congestions and save energy.

[1] E. Strubell, A. Ganesh and A. McCallum, “Energy and Policy Considerations for Deep Learning in NLP”, Annual Meeting of the Association for Computational Linguistics (ACL short). Florence, Italy. July 2019.

 

Goal

The objective of this line is to devise energy-aware computation algorithms, combining them with an intelligent network management, accounting also for distributed energy harvesting and storage hardware. To attain this goal, we envisage research activities centered around the following topics:

    • Sustainable computing for networking: Low-complex and energy-aware learning methods enabling network resource optimization and intelligent decision making (e.g., distributed and multi-agent reinforcement learning),
    • Sustainable learning for edge nodes: Methods enabling training and inference of AI models at the edge nodes (e.g., federated learning, gossip learning),
    • Sustainable computing for non-ICT systems: distributed and energy-aware learning methods to improve sustainability of other exogenous processes (e.g., urban population mobility, urban energy maps, air and sea water quality).

 

Methodology

This research line deals with those problems related to generation, communication, storage, and processing of information gathered from distributed and heterogeneous sources. The final objective is to extract knowledge from data and use it for optimal control. Therefore, particular attention is given to low-complex and energy-aware algorithm design, distributed optimization, data collection systems, information processing and computing platform architectures.

 

Publications

PhD Thesis

Courses and Seminars

  • P. Dini, Machine Learning for Mobile Communication Systems, course for the PhD School of Information Engineering department, Università degli Studi di Padova, June-July 2021.
  • P. Dini, Machine Learning for Mobile Communication Systems, course for the PhD School of Information Engineering department, Università degli Studi di Padova, May-June 2020.
  • P. Dini, Identifying Urban Anomalies through Mobile Data Processing, seminar at the Italian Networking Workshop 2020, Cavalese (Italy).
  • M. Miozzo, Self-Organized Solutions for 5G Networks with Recurrent Neural Networks, seminar at the Italian Networking Workshop 2020, Cavalese (Italy).
  • P. Dini, Data-driven Modeling and Optimization: a Networking Perspective, seminar at Politecnico di Bari, Italy, June 2019 [invited]
  • P. Dini, Fast Mobile Traffic Prediction from Raw Data using LSTM Networks, seminar at the Italian Networking Workshop 2019, Bromio (Italy).
  • P. Dini, Mobile Traffic Modeling by Intercepting Raw Data in the Radio Interface, seminar at the Italian Networking Workshop 2018, Courmayer (Italy).
  • P.Dini, Energy Sustainable Architecture and Methods for Future Mobile Communication Networks, seminar at the Summer School on Information Engineering 2017, Brixen (Italy). [invited]
  • P. Dini, A reinforcement learning approach to control renewable energy spending in energy harvesting cellular networks, seminar at the Italian Networking Workshop 2017, Falcade (Italy) [invited]
  • P. Dini, Sustainable Mobile Networks: the SCAVENGE vision, seminar at the Summer School on Information Engineering 2016, Brixen (Italy). [invited]

Thesis and Internships

  • L. A. Lopez Sobrado, Integration of heterogeneous data in a geographic information system for urban mobility analysis, Master thesis at Universitat Oberta de Catalunya, supervisor: Paolo Dini, January 2021 (in Spanish).
  • M. Canil, LTE Traffic Classification via Semi-Supervised Learning, internship for master thesis at Università di Padova, supervisor: Paolo Dini, Director: Michele Rossi, Padova, September 2020.
  • A. Pelati, Introducing Machine Learning Solutions For Modern Mobile Network Management, internship for master thesis at Politecnico di Torino, supervisor: Paolo Dini, Director: Michela Meo, Torino, January 2020.
  • P. Lastella, Implementation of prediction algorithms using Python, internship for Master Thesis at Politecnico di Bari, July-September 2019.
  • A. El Amine, Reinforcement Learning for Multi-Sleeping Control of Delay-Constrained and Energy-Optimal Small Base Stations, PhD Internship IMT-Atlantique – CTTC, May 2019.
  • A. Fernandez Gambín, Mobile traffic load classification through deep learning, director: P. Dini, advisor: M. Rossi, EU-MSCA Secondment, University of Padova-CTTC, April 2018-February 2019.
  • E. Cuoccio, Reinforcement learning for energy harvesting 5G mobile networks, director: P. Dini, advisors: A. Pascual and G. Boggia, joint Master Thesis by Universitat Politecnica de Catalunya and Politecnico di Bari, July 2016.
  • A. Dudnikova, Multi-criteria decision analysis for sleep mode in 5G networks, advisors: L. Giupponi, P. Dini, PhD internship, Università degli studi di Catania, February-September 2015.

If you are interested in working with us (as a trainee, PhD student or post-doc) please send an email with your CV and a tentative research proposal.

Funding

The activities of this research line are inserted within the following projects: