Stefano Mariani

Stefano Mariani

Overview

If you like programming, machine learning, data analysis, tinkering with Internet of Things devices, or mess around with networked computers, you’re in the right place :)

Here below you can find a list of proposed “activities”, as well as a record of concluded thesis and internships. Each ctivity may become a thesis or internship depending on the that goal we, together, decide to achieve, and on the expected effort required to achieve it. Activities are roughly divided in categories, representing the research topic they are meant to deal with: if you find an appealing topic, but not a satisfactory proposal therein, contact me and we’ll discuss your own proposals :) Finally, the numbers you see both in categories and proposals of activities therein are not a ranking of importance or a priority score, but merely a reference number to ease communications.

If you find something appealing to you, here is the “protocol” that I usually follow for setting up the thesis or internship:

  1. You contact me to tell me which categories or activities you find appealing
  2. We set up a quick meeting (either virtual or in presence) where I better describe the topic or proposal, giving you an overview and some reference examples of activities
  3. If you are still interested, I send you some reading material to give you an even better idea of the tasks to carry out as part of the proposed activity
  4. If you are still interested, you contact me to formally set up the internship or thesis

Until step (4), you are not expected to finish the activity: it’s just you trying to understand if you will like such activity as thesis or internship :)

1) Reinforcement learning in multiagent systems

Machine Learning (ML) is now pervasive in our everyday lives (Google translate, Siri speech recognition, etc.) and successfully applied in many application domains where a software (an agent) needs to learn to do something. However, modern software is rarely a monolithic entity that does everything by itself, rather a collection of indipendent services that collaborate to carry out a given task (e.g. Siri is connected to your Calendar, weather service, apps, etc.). Modern software is thus usually a multiagent system where multiple software entities need to coordinate to achieve a given goal.

Various forms of learning are used in multiagent systems (MAS) to let agents individually learn to accomplish a task, or to recognise a given state of the world. Instead, letting agents learn how to interact socially is yet underexplored in scientific literature, but a promising topic to let multiagent systems self-organise to achieve a shared or system goal.

The following activities unfold within this theme.

Activities

  1. apply reinforcement learning to NetLogo multi-agent system models for self-organisation such as slime mold aggregation, ant foraging, prey-predator. The goal is to let agents learn to communicate / cooperate / coordinate to achieve a shared goal.
  2. implement such environments in Python by using PyGame to avoid the limitations of NetLogo
  3. apply curriculum learning (read only the intro) in the same models
  4. apply transfer learning (just get the idea) in the same models
  5. apply reinforcement learning to cooperative driving scenarios (such as intersection crossing by self-driving vehicles) in custom simulators such as highway-env, Flow, SUMO-RL, or MetaDrive. The goal is to let agents (vehicles, or traffic lights, or intersection controllers) learn how to drive in traffic.

  6. adapt existing multi-agent learning environments to learn how to communicate / cooperate / coordinate to achieve better results. Examplary environments (like “games”):

Concluded

  • Luca Tusini. “Apprendimento per rinforzo di tecniche di comunicazione mediata per il coordinamento di sciami artificiali.” Bachelor degree in Management Engineering, 2021/2022.
  • Riccardo Santi. “Esperimenti di transfer learning per reti neurali con robot mobili.” Bachelor degree in Management Engineering, 2021/2022
  • Matteo Sigolotto. “Apprendimento autonomo in sistemi IoT mediante Reti Bayesiane.” Master degree in Computer Science Engineering, 2018/2019

2) Causal discovery and inference in the Internet of Things

Machine learning and deep learning techniques are nowadays pervasively used in everyday applications and devices (e.g. on smartphones) for their ability to accurately recognise patterns of associations between data. However, this ability also is one their limits: they can unveil associations between data, but can’t tell if such associations are in a cause-effect relationship (generally speaking).

Causal discovery and inference, instead, have exactly this ability to distinguish causation from correlation, hence are mostly used in critical domains such as healthcare, plant automation control, fault diagnosis.

The following activities unfold within this theme.

Activities:

  1. apply causal discovery algorithms to learn causal models of a domain of interest, such as smart homes (how sensors and actuators affect each other)
  2. apply Bayesian structure learning to do the same
  3. play with the causal discovery algorithms available here on the datasets available there. The goal is to assess both algorithms (for accuracy and performance) and the datasets (ease of usage, completeness, etc.)
  4. play with causal models / Bayesian networks libraries (such as CausalNex, Benchpress, and YLearn to assess their abilities (e.g. support to learning, performance of inference algorithms, etc.)
  5. extend any given causal discovery algorithm to a distributed setting, where data available for learning is not gathered at a single point, but is scattered across many networked databases

Concluded

  • Pasquale Roseti. “Multi-agent Learning of Causal Models in a Smart Home Environment.” Master degree in Computer Science Engineering, 2021/2022.

3) Coordination of vehicles for urban traffic management

Under construction

Being stuck in traffic is not funny for anyone…while self-driving cars could relieve us from the duty of daily driving, cooperative driving is necessary to enable self-driving cars to handle complex situations like crossing intersections, that require the coordination of multiple vehicles to establish a safe crossing order. Urban traffic management is thus an abundant source of coordination problems that future autonomous and connected vehicles will need to tackle to successfully hit the streets. The following activities unfold in this domain.

Thesis

  1. experiment with existing agent-based simulators adopted/adapted in traffic management domain
  2. design and implement an agent-based microscopic traffic simulator

Internships

  1. experiment with the SUMO traffic simulator to implement state of the art intersection crossing strategies
  2. develop a Graphical User Interface (GUI) for a simple simulator developed by graduate student Dario Ferrari

Concluded

  • Luca Bordini. “Multiagent Reinforcement Learning Strategies for Intersection Networks” Master degree in Computer Science Engineering, 2021/2022.
  • Dario Ferrari. “Coordinamento di veicoli autonomi basato su tecniche di argimentazione: simulatore ed esperimenti.” Bachelor degree in Management Engineering, 2021/2022.
  • Nicholas Glorio. “Strategie per la gestione delle intersezioni in presenza di veicoli autonomi.” Master degree in Computer Science Engineering, 2020/2021.
  • Marco Gambelli. “Gestione di incroci stradali per veicoli a guida autonoma basata su algoritmi a prenotazione e ad asta.” Master degree in Computer Science Engineering, 2020/2021.
  • Andrea Vitali. “Esperimenti di coordinazione veicolare tramite blockchain.” Master degree in Computer Science Engineering, 2019/2020.
  • Enrico Rossini. “Attraversamento di incroci per auto a guida autonoma: protocolli e simulazione.” Bachelor degree in Management Engineering, 2019/2020.

4) Argumentation protocols for joint deliberation/action and situation recognition in multiagent systems and IoT

Under construction

Many daily activities we carry out are assisted by software (e.g. digital assistants like Siri, Alexa, etc.), that (1) exploit artificial intelligence techniques, e.g. Machine Learning (ML), to deliver their services, and (2) work together with many other software entities to accomplish tasks (e.g. Siri is connected to your Calendar, weather service, apps, etc.).

Using ML may result in opaque systems that re not easily comprehended in their decision making by the user (e.g. why is Netflix suggesting that?). Moreover, having systems composed by multiple software pieces raises the problem of establishing how these pieces should coordinate to ahieve the desired goal. Many techniques exist to make ML explainable, that is able to otivate their decisions to humans, and many other exist to define coordination protocols amongst independent software pieces dictating how each piece should behave.

Computational argumentation is a promising research area that could potentially solve both problems at once.

Thesis

  1. develop a software framework for practical argumentation
  2. develop a software framework for playing dialogue games for multiagent negotiation and conflict resolution

Internships

  1. surveying commonsense reasoning state of art
  2. experiment with the Tweety Java library for computational argumentation
  3. experiment with the ArgTuProlog system for computational argumentation

Concluded

  • Dario Ferrari. “Coordinamento di veicoli autonomi basato su tecniche di argimentazione: simulatore ed esperimenti.” Bachelor degree in Management Engineering, 2021/2022.
  • Andrea Bicego. “Progettazione di Servizi IoT basati su Protocolli di Argomentazione.” Master degree in Management Engineering, 2017/2018.

5) Beyond AutoML: software engineering applied to machine learning pipelines

Under construction

Thesis

  1. design of a modular machine learning pipeline for predicting health outcomes in COPD/asthma patients

Internships

  1. using PMML/PFA for sharing machine learning models/pipelines across software platforms (e.g. R and Python)
  2. surveying AutoML state of art (e.g. open source libraries)
  3. serving machine learning models over the web

Concluded

  • Gabriele Rinaldi. “Sviluppo di un’applicazione web per la fruizione di modelli predittivi.” Bachelor degree in Management Engineering, 2021/2022.
  • Benedetta Becchi. “Sperimentazione della piattaforma KNIME: creazione di una pipeline di machine learning per il trattamento di dati sanitari.” Bachelor degree in Management Engineering, 2019/2020.

Miscellaneous (e.g. students’ proposals)

Concluded

  • Angelo Zerlenga. “Sviluppo di una webapp per l’assistenza alla mobilità elettrica: R(e)CHARGE” Bachelor degree in Management Engineering, 2021/2022.
  • Maicol Spaggiari. “BITCOIN: TRA TECNOLOGIA E RIVOLUZIONE MONETARIA.” Bachelor degree in Management Engineering, 2021/2022.
  • Riccardo Rivi. “Analisi predittiva di indici azionari: confronto tra modelli nel caso di S&P 500” Bachelor degree in Management Engineering, 2021/2022.
  • Gianluca Milano. “Identità digitale: la tecnologia Blockchain per documenti e certificazioni digitali” Master degree in Management Engineering, 2021/2022.
  • Giacomo Nasi. “Digital Twin applicati all’ IoT: esperimenti con la piattaforma OneM2M.” Bachelor degree in Management Engineering, 2021/2022.
  • Carmen Tomaiuolo. “La tecnologia EDI come strumento di integrazione: definizione, funzioni, il caso aziendale.” Bachelor degree in Management Engineering, 2021/2022.
  • Salvatore La Pica. “Utilizzo della Blockchain e degli Smart Contract in ambito risorse umane e supply chain” Bachelor degree in Management Engineering, 2020/2021.
  • Luca Morini. “Non SprECO: la web app per monitorare il proprio impatto ambientale quotidiano.” Bachelor degree in Management Engineering, 2020/2021.
  • Nicola Romano. “Finanza decentralizzata: dalla tecnologia Blockchain ad una nuova finanza.” Master degree in Management Engineering, 2020/2021.
  • Carla Marangi. “La sicurezza informatica ai tempi del COVID: rischi e misure protettive per lo smart working.” Bachelor degree in Management Engineering, 2019/2020.
  • Alessandra Occhionero. “Analisi della piattaforma AWS IoT Core: gestione di device IoT simulati.” Bachelor degree in Management Engineering, 2019/2020.
  • Luca Bartolini. “Digital Twin per industria 4.0: piattaforme e scenari applicativi.” Bachelor degree in Management Engineering, 2019/2020.
  • Matteo Santacaterina. “Analisi dati GeoJson a supporto di processi clinici: il Mapping DSS in CONNECARE.” Bachelor degree in Management Engineering, 2018/2019.
  • Andrea Canepari. “Geolocalizzazione e stratificazione di pazienti critici su mappa: il caso del Mapping DSS in CONNECARE.” Bachelor degree in Management Engineering, 2018/2019.
  • Jacopo Stefani. “SISTEMA DI SUPPORTO DECISIONALE IN AMBITO CLINICO: VISUALIZZAZIONE DEI PAZIENTI SU MAPPA.” Bachelor degree in Management Engineering, 2017/2018.
  • Andrea Alberini. “Applicazione Java per l’ottimizzazione dell’attività fisica tramite messaggi motivazionali contestualizzati.” Bachelor degree in Management Engineering, 2017/2018.

See also the legacy Unibo page for older thesis.

See also the legacy Unibo page for older projects / internships.