Skander Moalla

Room 203

Robert Hooke Building

University of Oxford.

Hi. I’m studying for an MSc in Advanced Computer Science at the University of Oxford. My research interests are in Reinforcement Learning (RL) and I am currently working on Multi-Agent RL at the Whiteson Research Lab (WhiRL) for my MSc thesis supervised by Mingfei Sun and Prof. Shimon Whiteson.

I previously interned as an applied scientist at Quincus where I built RL models to optimise middle-mile logistics, and as a software engineer at Amazon where I built an AI-based tool to revise business metrics across Amazon Transportation Services.

I graduated with a BSc in Mathematics and Computer Science from Ecole Polytechnique and was a visiting student at EPFL, the University of Toronto, and Stanford University.

At Oxford, I lead the Careers team of the Oxford Artificial Intelligence society (OxAI) and play as a libero for the men’s Blues volleyball team.

Interests

Reinforcement Learning Deep RL, Multi-Agent RL

Machine Learning

Education

Computer Science Oxford (MSc)

Maths & Computer Science Ecole Polytechnique (BSc)

Software Eng U of Toronto

AI & Entrepreneurship Stanford

Experience

RL Applied Scientist Quincus

SDE/SWE & ML Amazon

Neuroscience Blue Brain Project (EPFL)

News

Jun 12, 2022	I’ve been admitted with a fellowship to the PhD Program in Computer and Communication Sciences (EDIC) at EPFL and will start the program in september 2022.

Selected Publications

BSc Thesis

A Multidisciplinary Analysis of the Link between Structure and Function in Biological Neural Networks

Moalla, Skander

Apr 2020

Abs HTML PDF

In recent applications of algebraic topology to explore and analyze digital reconstructions of neocortical microcircuitry developed at the Blue Brain Project, researchers have been able to quantify the structural complexity of a neural network and its activity. Those results suggest a correlation between the structural complexity and the functional complexity of a neural network. In this report, we first provide an overview of the multidisciplinary approach one has to adopt to study such a correlation. Incidentally, we build the simple framework of choice: an implementation of a simple yet very efficient neuron model using Brian — a user-friendly python simulator for spiking neurons. Then, we focus on the topological structures suspected to drive this correlation, namely, directed cliques of neurons and build a network where this correlation can be measured and analyzed using tools from information theory. In particular, we look at the evolution of mutual information between pairs of neurons within cliques and break this measure down using the partial information decomposition. The results of our analysis suggest that when subject to a stimulus, high dimensional cliques transform the information coming from the stimulus to generate new information reflecting their high degree of organization.