Open positions
PhD Positions at MaLGa
Type
PhD student position
Supervisors

All MaLGa faculty
About
We are offering several PhD positions, starting in Fall 2023, on all the research themes explored at MaLGa, from the mathematics of machine learning, optimisation, signal processing and inverse problems, to the applications of machine learning to vision, health, physics and biology.
If you are interested you should fill in the expression of interest below, selecting the MaLGa unit(s) whose research themes are of interest to you, and/or one (or more) of the following projects, which represent a nonexhaustive list of possible PhD topics.
Perspective candidates should make an expression of interest before 26/02/2023.
1. Inverse problems via compressed sensing and machine learning. Supervised by G. S. Alberti. (CHarML)
This project is focused on using methods of applied harmonic analysis, in particular compressed sensing, wavelet theory and approximation theory, and of machine learning to solve inverse problems. We shall focus on inverse problems described by integral operators, as well as those modelled by PDEs.
2. Machine learning methods for imaging and vision. Supervised by F. Odone, N. Noceti and V. P. Pastore. (MLV)
We have different projects focused on the study, design and development of computer vision methods leveraging machine learning. Possible research directions include: Machine Learning for biomedical imaging, Causality and learning for Computer Vision, Attention mechanisms for motion understanding.
3. Physics informed algorithms for sensing and navigating turbulence. Supervised by A. Seminara and A. Verri. (PiMLB)
Living systems collect information from their environment and make crucial decisions to e.g. reach food and mates or escape danger. The physics of fluids dictates that the mechanical and chemical cues emitted by the target are dramatically noisy, which makes it hard to navigate. We use reinforcement learning to model decision making of living organisms informed by sensory cues carried by air and water. We merge theory, numerical simulations and experiments with living organisms to understand the fundamental constraints that shape decision making in uncertain environments.
4. Machine learning methods for structured data. Supervised by A. Barla. (LCSL)
The project will investigate methods and best practices for structured, temporal, and multimodal data. The research themes include robust, reproducible and interpretable machine and deep learning methods for structured data. We will particularly focus on applications that exploit textual data and data representation through graphs.
5. HPC and efficient large scale ML (Leonardo Labs). Supervised by L. Rosasco. (LCSL)
The success of ML approaches based on massive models is contrasted with an ever increasing demand for computation resources. The development of novel efficient solutions is an important scientific question, but also an urgent practical need. In this project, we aim at combining and blend techniques from HPC and ML to design algorithms tailored to the available computational architectures, as well as devise computational schemes fully adapted to the best ML solution for the problem at hand. The ideal candidate should have strong numerical and computational skills, with an interest to interact with theoreticians and engage in applications. The position is joint with Leonardo Labs and in particular with Carlo Cavazzoni.
6. Provably efficient and trustworthy ML (Leonardo Labs). Supervised by L. Rosasco. (LCSL)
As ML si deployed in sensitive scenarios, it becomes crucial to be able to rigorously certify its properties. This project aims at deriving theoretical guarantees in terms of accuracy as well as computational costs. We will strive to characterise the efficiency of state of the art machine learning algorithms, while taking hint towards new improved solutions. We are looking for a candidate with strong mathematical and computational skills, interested in deriving novel theoretical results with direct practical implications. The position is joint with Leonardo Labs and in particular with Alessandro Nicolosi.
7. Algorithmic regularization for modern machine learning. Supervised by L. Rosasco, S. Villa, S. Di Marino. (LCSL)
Recent machine learning advances show that computations can be used to drive the inductive bias of large overparameterized learning models. This project aims at understanding the foundations of algorithmic regularization to develop sound and efficient largescale learning systems. The project will blend ideas from optimization, statistics and inverse problems. While theoretical in nature, the project will strive to derive practical solutions. The ideal candidate should have excellent mathematical and computational skills.
8. Efficient and physicsinformed machine learning. Supervised by L. Rosasco. (LCSL)
This project aims at developing provably trustworthy and efficient algorithms for modern machine learning applications, with particular emphasis on infinite dimensional problems defined by integral and partial differential equations. A possible focus will be on the study of machine learning solutions for illposed inverse problems. Another direction will be the study of dynamical systems. The goal is to develop approaches for which rigorous guarantees can be developed both in terms of accuracy and computational requirements. The ideal candidate should have an excellent mathematical background.
9. Efficient zeroth order optimization for ML. Supervised by S. Villa, S. Di Marino, L. Rosasco. (LCSL)
In some machine learning applications it is necessary to minimise a function for which no analytical form is readily available, and only a zerothorder blackbox or a simulation oracle giving the function value at a given point is accessible. In this context, only zeroth order optimization methods can be used. We will focus on finite difference and consensus based optimization methods. In this project we will try to go beyond the state of the art by relaxing the assumptions under which convergence can be established, while aiming to prove more quantitative results, directly related to practical efficiency. An excellent mathematical background and computational skills are required.
Perspective candidates should make an expression of interest before 26/02/2023.
Postdoc in EthicalAI Methods for Structured Data
Type
Postdoc position
Supervisor
About
We are looking for one postdoc in EthicalAI methods and best practices for structured, temporal, and multimodal data.
The research themes include robust, reproducible and interpretable machine and deep learning methods for structured data. We will particularly focus on applications that exploit textual data and data representation through graphs. A prior experience in machine learning is more than welcome.
The position starts on March 1, 2023, with some flexibility. The duration of the contract is 1 year. The postdoc salary is ~27keur/year before taxes.
This position is funded by the PNRR project “Robotics and AI for Socioeconomic Empowerment  RAISE”.
Deadline for application is January, 23 2023.
To submit your application follow the link to the application system.
After submitting the application or for further information, send an email to annalisa.barla@unige.it.
Postdoc in Machine Learning for Inverse Problems
Type
Postdoc position
Supervisors
About
The project focuses on machine learning approaches for illposed inverse problems with an emphasis on infinitedimensional problems defined by integral and partial differential equations. The goal is to develop algorithms for which rigorous guarantees can be developed both in terms of accuracy, stability and computational requirements. As such, tools from inverse problems, statistics and optimization will be combined. The ideal candidate should have an excellent mathematical background, in particular in functional analysis.
The position starts as soon as possible, but this is flexible. The duration of the contract is 1 year, renewable for up to 2 years. The postdoc salary varies depending on experience and is commensurate to international standards.
Perspective candidates should make an expression of interest before 26/02/2023.