Giulia Clerici

Machine Learning Researcher + Scientific coordinator

About Me

I am a Machine Learning Researcher at Università degli Studi di Milano and the Scientific Coordinator of the ELLIS Unit Milan as well as of the ELLIS Fellowship Program, which is part of the "European Lighthouse of AI for Sustainability" (ELIAS) Project.
I received my Ph.D. in Computer Science in April 2024 at Università degli Studi di Milano, focusing my research on developing new multiarmed bandits models to address personalized non-stationary behaviours in Music Recommender Systems.

Resume

Here is my education, work experience, & some skills I've got.

Experience

AI researcher, Scientific coordinator

Università degli Studi di Milano
ELLIS - European Laboratory for Learning and Intelligent Systems

February 2023 - Present

I'm currently conducting research on Linear Non-stationary Bandits and working as a Scientific coordinator of the ELLIS Unit Milan as well as of the ELLIS Fellowship Program for the European Project ELIAS - European Lighthouse of AI for Sustainability.

Teaching Assistant for Machine Learning

Università degli Studi di Milano

October 2021 - January 2024

Teaching Assistant for the course Statistical Methods for Machine Learning, held by Prof. Nicolò Cesa-Bianchi. Evaluation of 200+ experimental projects on Neural Networks and examination of 50+ students.

Tutor for Basics of Computer Science, Student collaborator

Università degli Studi di Milano

October 2017 - July 2020

I tutored students of Medicine and Dentistry on the basics of Computer Science, delivering theoretical lectures and labs. Management of school-work alternation activities and orientation events for the Computer Science Department.

Web and Database Developer

Social Thingum

April 2017 - July 2017

Between my BSc and MSc, I worked as a Back End Developer, specifically as a Web and Database Developer using CakePHP, SQL, HTML, CSS.

Education

Ph.D. in Computer Science

Università degli Studi di Milano

November 2020 - April 2024

I received a Ph.D. (doctoral degree) in Computer Science. During my Ph.D., my research interests focused on Machine Learning and Multiarmed Bandits. Specifically, I focused on non-stationary Multiarmed Bandits for satiation and seasonality phenomena in Music Recommender Systems.

Summer School on Online Learning

Sapienza Università di Roma

September 2021

European Summer School on Learning in Games, Markets, and Online Decision Making.

Master of Science in Computer Science

Università degli Studi di Milano

October 2017 - June 2020

My Master thesis concerned real-time audio synthesis based on classification of environmental sounds for the creation of a smart drum kit using Bela.

Bachelor of Science in Computer Science

Università degli Studi di Milano

October 2013 - April 2017

During my Bachelor thesis I worked on a prototype for text-independent automatic speaker and singer identification.

Skills & Expertise

  • Multiarmed bandits
  • Machine Learning
  • Scientific research
  • Project management
  • Presentation skills
  • Python and ML libraries
  • MatLab
  • C
  • Java
  • SQL
  • PHP
  • HTML
  • CSS

Volunteering experiences

Local chair

ALT - Algorithmic Learning Theory 2025

Milan, July 2024 - February 2025

I'm serving as Local Chair for the Algorithmic Learning Theory (ALT) conference which will take place in Milan from February 24th to February 27th. This job involves coordinating all on-site logistics, including venue arrangements, technical support, and local staffing, to ensure the smooth execution of the event.

Super Volunteer

WiML - Women in Machine Learning @ NeurIPS 2023

New Orleans, December 2023

I served as a (in-person) Super Volunteer for the WiML workshop at NeurIPS 2023 in New Orleans, LA, USA. My role was to coordinate other volunteers for some of the sessions (e.g. talks, panel) of the workshop.

Breakout Session and Logistics Co-Chair

WiML - Women in Machine Learning @ ICML 2022

Baltimore, April 2022 - July 2022

I served as Breakout Program and Logistics Co-Chair for the Women In Machine Learning Un-Workshop at the International Conference in Machine Learning (ICML) 2022 (Baltimore, MD, USA). My work consisted in organizing and planning part of the event in the preceding months and ensuring a smooth experience of the event, which was hybrid, both in-person and virtual. I was primarily responsible for the logistics, planning of the breakout sessions, selecting proposals, contacting possible presenters, coordinating with volunteers, organizers and staff, and structuring day-of-event volunteers' tasks. In addition, I was responsible for the in-person and virtual communication and coordination with the students involved in the breakout sessions.

Volunteer Staff

STOC - Symposium on Theory of Computing 2022

June 2022

As a Ph.D. student, I volunteered for the STOC conference 2022 (Rome, Italy) to assist the session chairs during the oral presentations, helping with the equipment, communicating with the presenters, and ensuring a smooth fruition of the in-person talks to the virtual audience.

Volunteer Staff

LIM - Laboratorio di Informatica Musicale

June 2018 - May 2019

I was a part of the volunteer staff of the Laboratorio di Informatica Musicale, Università degli Studi di Milano for two consecutive years, in 2018 and 2019, for the event "FIM Salone della Formazione e dell'Innovazione Musicale" (Milan, Italy). In this role I was illustrating and explaining the lab's technologies and projects to a diverse audience: from industry people, to students, and general public.

Volunteer Staff

LIM - Laboratorio di Informatica Musicale

January 2019

As a Master's student, I volunteered for the MMRP conference in 2019 (Milan, Italy) by helping with the registration of attendees and the day-of-event logistics. I assisted session chairs during the oral presentations, helping with the equipment, communicating with the presenters, and ensuring a smooth fruition of the in-person talks to the virtual audience.

Scientific Publications

Scientific Publications in Conferences.

PhD Thesis Link to the Thesis

Non-Stationary Multiarmed Bandits for Satiation and Seasonality Phenomena in Music Recommender Systems

This thesis delves into theoretical aspects of non-stationary multiarmed bandits, motivated by their application to music recommender systems. An intrinsic challenge of such systems lies in evolving user preferences. Rather than finding a single optimal item, the objective is to craft an ordered sequence of items aligning with the user's behaviour. Indeed, these systems need to address dynamic user preferences, characterized by phenomena like satiation. Our goal is to study these phenomena in a sequential learning setting characterized by partial feedback, adopting multiarmed bandits as the approach to tackle this problem. We first introduce a novel model with finitely many arms, which handles contrasting phenomena like satiation and seasonality in a unified framework. We formalize this model, called Last Switch Dependent Bandit, as a non-stationary multiarmed bandit where the expected reward of an arm is determined by its state, which indicates the time elapsed since the arm last took part in a switch of actions. This model can generalize to different types of non-stationarity as we relaxed many typical assumptions. Furthermore, it can recover state-dependent bandits in the literature. In this thesis, we will discuss this bandit problem and the solution proposed to solve it, which is based on upper confidence bounds and techniques derived from combinatorial semi-bandits. We conclude this work by providing an upper bound of the regret, to assess the performance of the proposed solution. Aware of the limitations of finite action sets, which are not always representative of real-world applications, a new linear bandit model is proposed to handle non-stationary phenomena within an infinite set of actions, posing complex challenges for cross-arm dependencies. We formalize a model called Linear Bandit with Memory, where the expected reward of an action is influenced by the learner's past actions in a fixed-size window, called memory matrix. Specifically, the two parameters $m$ and $\gamma$, the size of the window and the exponent respectively, characterizing this memory matrix can produce two types of behaviours: rotting and rising. In our discussion, we show how our model generalizes stationary linear bandits, as well as partially recovering rested rotting and rising bandits in the limit m -> + infinity. We study the complex problem of modelling the interactions among arms in a linear non-stationary setting and propose a solution to solve it. Furthermore, we study the setting where m and gamma are unknown and propose a meta-bandit algorithm for model selection to jointly learn the parameters and solve the bandit problem. For both models, our contributions consist in defining novel multiarmed bandit problems, proving sound theoretical guarantees, and discussing our contributions with respect to the literature. Additionally, within the context of music streaming services, we present an additional line of research exploring the Spotify\texttrademark music credits network. We represent this music credits network as a directed graph and study the relationship between music genres and graph-related metrics. After observing interesting patterns on several centrality measures conditioned on music genre, we introduce a novel node-wise index of reciprocity as a potential index for informed recommendations.

Scientific Conference Link to the paper

Linear Bandits with Memory

Nonstationary phenomena, such as satiation effects in recommendations, have mostly been modeled using bandits with finitely many arms. However, the richer action space provided by linear bandits is often preferred in practice. In this work, we introduce a novel nonstationary linear bandit model, where current rewards are influenced by the learner’s past actions in a fixed-size window. Our model, which recovers stationary linear bandits as a special case, leverages two parameters: the window size m ≥ 0, and an exponent γ that captures the rotting (γ < 0) or rising (γ > 0) nature of the phenomenon. When both m and γ are known, we propose and analyze a variant of OFUL which minimizes regret against cyclic policies. By choosing the cycle length so as to trade-off approximation and estimation errors, we then prove a bound of order √d (m + 1) 1 2 +max{γ,0} T^3/4 (ignoring log factors) on the regret against the optimal sequence of actions, where T is the horizon and d is the dimension of the linear action space. Through a bandit model selection approach, our results are then extended to the case where both m and γ are unknown. Finally, we complement our theoretical results with experiments comparing our approach to natural baselines.

Scientific Conference Link to the paper

Citation is not Collaboration: Music-Genre Dependence of Graph-Related Metrics in a Music Credits Network

We present a study of the relationship between music genres and graph-related metrics in a directed graph of music credits built using data from Spotify. Our objective is to examine crediting patterns and their dependence on music genre and artist popularity. To this end, we introduce a node-wise index of reciprocity, which could be a useful feature in recommendation systems. We argue that reciprocity allows distinguishing between the two types of connections: citations and collaborations. Previous works analyse only undirected graphs of credits, making the assumption that every credit implies a collaboration. However, this discards all information about reciprocity. To avoid this oversimplification, we define a directed graph. We show that, as previously found, the most central artists in the network are classical and hip-hop artists. Then, we analyse the reciprocity of artists to demonstrate that the high centrality of the two groups is the result of two different phenomena. Classical artists have low reciprocity and most of their connections are attributable to citations, while hip-hop artists have high reciprocity and most of their connections are true collaborations.

Scientific Conference Link to the paper

A Last Switch Dependent Analysis of Satiation and Seasonality in Bandits

Motivated by the fact that humans like some level of unpredictability or novelty, and might therefore get quickly bored when interacting with a stationary policy, we introduce a novel non-stationary bandit problem, where the expected reward of an arm is fully determined by the time elapsed since the arm last took part in a switch of actions. Our model generalizes previous notions of delay-dependent rewards, and also relaxes most assumptions on the reward function. This enables the modeling of phenomena such as progressive satiation and periodic behaviours. Building upon the Combinatorial Semi-Bandits (CSB) framework, we design an algorithm and prove a bound on its regret with respect to the optimal non-stationary policy (which is NP-hard to compute). Similarly to previous works, our regret analysis is based on defining and solving an appropriate trade-off between approximation and estimation. Preliminary experiments confirm the superiority of our algorithm over both the oracle greedy approach and a vanilla CSB solver.

Talks

Talks, Orals, Poster presentations at Conferences/Events.

Poster presentation

Next up! International Conference on Learning Representations ICLR 2025 - Singapore.

We're happy to announce that our paper Linear Bandits with Memory, a joint work with Pierre Laforgue and Nicolò Cesa-Bianchi, has been accepted at ICLR 2025, in Singapore!

Poster presentation

EWRL European Workshop on Reinforcement Learning 2024 - Toulouse, France.

In October 2024 I had the pleasure of attending EWRL in Toulouse, where I presented my poster on Linear Bandits with Memory.

Oral presentation

Presentation of the ELLIS Unit Milan at the ELLIS Coordinators' Meeting 2024 - Sofia, Bulgaria.

In September 2024 I joined other ELLIS Coordinators at the first ever ELLIS Coordinators' Meeting in Sofia, Bulgaria. During the workshop I had the chance to give a talk to present the ELLIS Unit Milan and its research activities and to connect with colleagues working in other ELLIS Units around Europe.

Oral talk and poster presentation at Scientific Conference

SMC Sound and Music Computing Conference 2023 - Stockholm, Sweden.

In June 2023 I had the great pleasure to give a talk and present a poster on a joint work with colleague Marco Tiraboschi on Citation is not Collaboration: Music-Genre Dependence of Graph-Related Metrics in a Music Credits Network.

Poster Presentation at Scientific Conference

EWRL European Workshop on Reinforcement Learning 2022 - Milan, Italy.

In September 2022 I presented my poster on A Last Switch Dependent Analysis of Satiation and Seasonality in Bandits, a joint work with Pierre Laforgue and Nicolò Cesa-Bianchi, at EWRL in Milan.

Online talk and poster presentation at Scientific Conference

AISTATS Artificial Intelligence and Statistics 2022 - Virtual Conference.

In March 2022 I presented my work on A Last Switch Dependent Analysis of Satiation and Seasonality in Bandits, a joint work with Pierre Laforgue and Nicolò Cesa-Bianchi, at the AISTATS virtual conference.

Scientific Conference

SpotiGem 2021 - Milan.

In October 2021 my colleague Marco Tiraboschi and I presented our work on Music-Genre Dependence of Graph-Related Metrics in a Music Collaboration Network in Milan, during the SpotiGen event, an interdisciplinary project between the Departments of Computer Science, Cultural Heritage and Social and Political Sciences.