Theses
If you are interested in Cloud, Edge or Fog Computing algorithms and applications and are willing to work on these topics, you can email me to book a meeting. The open topics are typically research-oriented, and they are:
- load balancing and distributed scheduling algorithms for Fog and Edge Computing:
- targeting energy (Green Edge Computing);
- QoS, performance metrics like latency/drop rate;
- Reinforcement Learning-based approaches for the previous point;
- offloading algorithms for mobile-based applications, targeting energy and latency;
- technologies oriented solutions for container orchestration;
- any other related topic that you want to deepen.
Generally, the theses can require programming skills in Python (especially for simulations), Go and technologies like Docker Containers, Kubernetes, and a minimum of Linux sysadmin skills for more practical tests. The experiments can be conducted even on our framework called P2PFaaS, which is already installed in a cluster of Raspberry Pi. We are also open to conducting more theoretical-oriented theses if you are interested. Also, see the list of Publications for a glance at my research interests and expertise.
Please don’t be scared by new technologies or concept that you don’t know, the role of the thesis is also getting involved into learning new things that you have never seen before and having fun doing that!
Open problems
This is a list of precise open problems that can be immediately deepened in a MSc thesis. Contact me for further details.
- Network selection with Reinforcement Learning at the Edge Explore the improvement that a RL-based network selector can reach in case of selecting multiple DNNs for inference with limited constraints in terms of power and inference time.
- Microcontroller-based inference at the Edge Try to develop a solution for remotely deploying DNNs in a microcontroller using TensorFlow Lite. We will supply you a brand new microcontroller for conducting the thesis.
- Reinforcement Learning based distributed scheduling for Green Computing Try to develop RL-based algorithm which is in charge of making scheduling decisions based on its current energy available and the further solar panel recharge. You can develop the algorithm in simulation and also on a real set of Raspberry Pi that we have in lab. The thesis is a continuation of this paper and this one.
Past Theses
This is a list of the past students that I assisted in their MSc thesis, some of the works have also been published in form of a scientific publication, in that case the link is attached:
- A Reinforment Learning approach for Optimal Neural Network Selection, Nicolas Antonio Benko, October 2024 (to be defended);
- Real-time and Energy-aware Scheduling for Edge-to-Cloud Continuum based on Reinforcement Learning, Andrea Panceri, June 2024;
- A Distributed and Cloud-Driven Framework for Edge AI Services Deployment and Energy-Aware Load Balancing, Matteo Feliziani, January 2024;
- A Performance Study of Cloud-backed Edge AI inference with AWS Greengrass and Kinesis, Onkarappa Belludi Siddesh, January 2024;
- TF Optimizer, user friendly software for neural network models optimization, Luca Giovannesi, October 2023. [PAPER];
- Shared Mobile AR Experience based on Edge/Fog Computing: Design and Performance, Hassaan Qureshi, October 2023.
- A Study on Energy Efficiency in Edge-assisted VR Applications with Meta Quest 2 for Disaster Management, Lorenzo Romagnoli, July 2023. [PAPER];
- Energy Balancing Algorithms for Green Edge Computing, Marco Ciancia, March 2023. [PAPER];
- A Latency-levelling Load Balancing Algorithm for Fog and Edge Computing, January 2023, Marco Magnani. [PAPER];
- A Double-step Reinforcement Learning Algorithm for Online Scheduling in Edge and Fog Computing, January 2023, Ahmed Fayez Moustafa Tayel [PAPER];
- Local and Remote Fog based Trade-offs for QOE in VR Applications by using CloudXR and Oculus Air Link, June 2022, Gabriele Maiorano. [PAPER];