Profile picture for user VRIZZI S

Stefano Vrizzi


PhD student

29 rue d'Ulm
75005 Paris France

Mathematics of Neural Circuits
Selected publications
International Journal article  

Lussange, J., Vrizzi, S., Bourgeois-Gironde, S., Palminteri, S. & Gutkin, B. (2022). Stock Price Formation: Precepts from a Multi-Agent Reinforcement Learning Model. Comput Econ. doi:10.1007/s10614-022-10249-3

Short bio


What is the impact of individual decisions on a collective system?


This is the fundamental question driving my career. To answer it, I pursued knowledge in two fields. On one side, I specialised in information processing at individual level; on the other side, I trained at the intersection between economics and sustainability, the key marker of human ability to manage resources and create value as a society.


Currently, my PhD project aims to better understand finance bottom-up: the way a collective stock price forms from individual trading strategies. More specifically, I am assessing which combination of cognitive traits, in a Multi-Agent Reinforcement Learning model, capture more closely the statistics of London Stock Exchange data.


Along scientific research, my primary effort is to encourage knowledge sharing through conversations. For this reason, I have been developing solid professional experience in conference management, while training in science communication and pursuing open access.


Further information on my background is available on the following tabs.

Information processing



My fundamental question is about the relation between individuals and society: how individual decisions shape collective outcomes. My interest in this question developed thanks to the high school programme in Italy, which mixed science, philosophy and international projects on human rights, such as the Model of United Nations.
I wanted to start understanding my question bottom-up, from its fundamentals, from a scientific perspective, so I learnt about the biological and physiological bases of information processing by graduating from my BSc in Neuroscience at the University of Leeds, UK. In order to model my questions, I started building a theoretical and computational background through BSc research experience, a research placement at the University of Washington, USA, and a MSc in Neural Information Processing at the International Max Planck Research School in Tübingen, Germany.


PHD [1]

Currently, I am extending a Multi-Agent Reinforcement Learning (MARL) model called SYMBA (SYstème Multiagent Boursier Artificiel), which simulates stock market trading. This model is novel in two key features: i) agents can learn trading strategies by reinforcement learning and ii) their orders are collected and managed by a centralised order book. In the 2022 paper, we show how it can study the time-dependency of policy performance and heterogeneity. After embedding cognitive traits in the learning algorithms of the agents, I am assessing the impact of the distribution of cognitive traits in market dynamics through computational simulations.



At individual level, I am interested in understanding how differently humans process information and make decisions; at a collective level, I am interested in the effect of this heterogeneity on society.

As we, as humans, interpret the surroundings through our nervous system, my first goal was to learn how to model neural activity. My research experience in computational neuroscience developed since my first year of BSc [2]. I started from a simpler, yet fascinating, circuitry: the spinal motor pathways. Here, behavioural measurements (e.g. electromyography of motor output) are less subject to interpretation than in cognition. A key milestone in my training in computational neuroscience came as undergraduate research fellow [3], where I extended a computational model to simulate the effects of artificially induced neural plasticity by a Brain Computer Spinal Cord Interface (BCSI).


I then wondered not only about how we can process information, but also how we cannot. There may be physical constraints imposing patterns and rules, which would determine the extent and possible types of heterogeneity of information processing across a population of information processing systems. Since the states accessible to a physical system are commonly dictated by energy constraints, I explored the relation between energy and information as physical quantities for my MSc thesis [7], comparing analytical predictions with computational results concerning the work extracted from a generalised Szilard engine. This project introduced me to the field of information thermodynamics and statistical physics.


After learning about the physiological components and physical constraints of information processing, my research scaled up to motor and behavioural tasks, while keeping a quantitative approach. I investigated heterogeneity in human information processing in three main projects:

- my BSc dissertation [4], quantifying distance between muscle activity patterns among human subjects, given the same motor task;

- my MSc lab rotation [5], classifying muscle activity patterns within-subject from high density recordings during daily motor tasks;

- my work as research assistant [6], analysing test-retest inter-individual variability in human reinforcement learning


In summary, my primary research experience has been based on computational models and machine learning, especially supervised learning and unsupervised learning, while my current PhD project focuses on Multi-Agent Reinforcement Learning simulations.




[1] Boris Gutkin, Group for Neural Theory, Département d'études cognitives, École normale supérieure, Paris, France
Stefano Palminteri, Human Reinforcement Learning Team, Laboratoire de Neurosciences Cognitives & Computationnelles, Département d'études cognitives, École normale supérieure, Paris, France
Damien Challet, Laboratoire de Mathématiques et Informatique pour la Complexité et les Systèmes, CentraleSupélec, Gif-sur-Yvette, France


[2] Samit Chakrabarty, School of Biomedical Sciences, Faculty of Biological Sciences, University of Leeds, Leeds, UK
Marc de Kamps, School of Computing, Faculty of Engineering and Physical Sciences, University of Leeds, Leeds, UK


[3] Guillaume Lajoie & Adrienne Fairhall, Fairhall Lab, University of Washington Institute for Neuroengineering, Seattle, USA


[4] Samit Chakrabarty, School of Biomedical Sciences, Faculty of Biological Sciences, University of Leeds, Leeds, UK


[5] Martin Giese, Section for Computational Sensomotorics, Hertie Institute for Clinical Brain Research, Centre for Integrative Neuroscience, Tübingen, Germany
Leonardo Gizzi, Neuromechanics Laboratory, University of Stuttgart, Stuttgart, Germany


[6] Stefano Palminteri, Human Reinforcement Learning Team, Laboratoire de Neurosciences Cognitives & Computationnelles, Département d'études cognitives, École normale supérieure, Paris, France
Maël Lebreton, Paris School of Economics, Paris, France


[7] Matteo Marsili & Edgar Roldan, Quantitative Life Science, International Centre for Theoretical Physics, Trieste, Italy
Anna Levina, Max Planck Institute for Biological Cybernetics, Tübingen, Germany


In addition to the outstanding supervision I had the honour to receive during all these research projects, a crucial building block of my critical thinking development came from periodical meetings with my BSc tutor Dr Ian Wood.

Economics & sustainability

There are two main fields dealing with the eco (from ancient Greek, oikos, 'house') we inhabit: eco-logy and eco-nomics. The former describes the eco, the latter norms its management. Society and its wellness therefore depend on the alignment between these two fields.


My experience on this intersection developed especially through my secondary research on eco-economic decoupling at the Global Trends Unit, within the European Parliamentary Research Service at the EU Parliament in Brussels, Belgium. Here I learnt about the socio-economic transition Deals that several governments envisage to implement to aim for a more sustainable society. The success or failure of these Deals fundamentally hinges upon the assumption that it is possible to achieve an eco-economic decoupling that is absolute, global, permanent, fast enough and valid across environmental indicators. I focused on climate change, namely the economics around the consumption of energy produced from fossil fuels. In addition to participating in summer schools on energy econometricstransformative economic policies and climate change, I also collaborated with Rethinking Economics at Sciences Po to popularise the debate on internalising carbon pricing.



Promoting knowledge

A thriving democracy relies on dialogue. For this reason, along science, my primary effort is to promote knowledge sharing.


Conference management
My experience in organising large events (600+ participants) began at high school, as student council representative. At university, in 2016, I condensed this experience to found Leeds Neural Networks, together with a group of peer students in Leeds, UK. As meaning arises from connections, this student society, still running today, aims to connect: connect pieces of knowledge and connect people. It organises formal and informal events to discuss neuroscience and related fields, mostly at the University of Leeds and in town. Side activities also concerned charity fundraising for brain tumour research and organising the quiz night for the British Neuroscience Association Festival.
This experience prepared me to manage the keynote speakers' invitations for the Global Scholars Symposium 2019 at the University of Oxford, and to work as the trainee team manager for the logistics of logistics ESPAS conference 2019 at the European Parliament.


Research training in AI & society
I recently coordinated an inter-institutional team between Scuola Superiore Sant'Anna (Pisa, Italy), Scuola Normale Superiore (Pisa, Italy) and my university, to organise a spring school named "Ethos+Tekhnè : a new generation of AI researchers". It took place in Pisa in March 2023, thanks to support of the European Engineering Learning Innovation and Science Alliance (EELISA).

The goal was to encourage the upcoming generation of AI developers to combine their tekhne with ethos.

The peculiarity of this initiative is its bottom-up organisation and its pluralism.

1) Bottom-up: it is organised by young researchers, PhD students, young professionals, across fields and institutions.
2) Plural: it is interdisciplinary, covering a wide range of topics and perspectives, with invited speakers encompassing scientists, big tech engineers, philosophers, economists, lawyers, policy advisors, Members of the European Parliament and journalists; moreover, is designed to offer a variety of activities, from lectures to workshops, coding tutorials, discussion panels and more.


Science communication
In the past, my involvement also concretised as short articles for the British Neuroscience Association, or by promoting neuroscience with the Human Brain Project, or as TEDx speaker. More recently, I facilitated workshops at OpenCon 2018 and Mozilla Festival 2019, leveraging open access to share knowledge. Finally, in 2021 I joined La Fresque du Climat to offer educational workshops on climate change at COP26, Glasgow, UK.


(2021-22) Statistique: probabilités (Travaux Dirigés), niveau L2, École d’économie de la Sorbonne, Université Paris 1, Paris, France




Previous funding