Sweden
A mathematical mind with solid foundation in physics, both machine learning engineer ("civilingenjör") from KTH Royal Institute of Technology, and generalist engineer with a major in biomedical engineering from Ecole Centrale Marseille. Currently PhD candidate in machine learning at KTH, and faculty member at DIS Study Abroad institution where I teach data science. I am very motivated by artificial intelligence and applied mathematics projects having a positive impact on the society, especially in the medical field! And I am also very interested in strategy consulting and entrepreneurship! My topics: representation learning, data analysis, hierarchical data, hyperbolic machine learning, computational geometry, topological data analysis, applications to biology and neurosciences, Brain-Computer Interfaces. My PhD is about hyperbolic data analysis with applications to biology and cognitive psychology. In short, I use some advanced mathematics (hyperbolic geometry) combined with machine learning to develop new methods that can better understand and analyze hierarchical data. I apply our methods on different datasets possessing such hierarchical structure, in particular in biology (single-cell RNA sequencing, protein evolution, phylogenetics) and in cognitive psychology (olfactory perception). Below is a quote for those who might think that PhDs in computer science would be very talented in repairing computers... "Computer Science is no more about computers than astronomy is about telescopes, biology is about microscopes or chemistry is about beakers and test tubes." From Fellows & Parberry in "SIGACT Trying to Get Children Excited About CS" So, what is computer science? Let's talk about it! Computers and mathematics are my tools; computation, information and automation my objects of study; artificial intelligence my speciality, data my material, biology and neurosciences my applications and inspirations. Married, based in Stockholm.
PhD in Machine Learning under supervision of Professor Danica Kragic Jensfelt at the Robotics Perception and Learning division of KTH Royal Institute of Technology. Our main job is to publish high quality scientific papers. This includes finding relevant projects, managing time with plans and deadlines, solving complex problems, programming, running experiments to test hypotheses and comparing our methods to the state of the art, scientific writing, rewriting and improving our papers based on peer reviews, presenting our work to other researchers/partners that are not necessarily from our field, travelling to international conferences, giving talks and poster presentations, reviewing other papers with critical thinking, teaching to master's students and PhD peers, supervising master theses... Topics: Geometric and topological methods for representation learning, hyperbolic data analysis of hierarchical data, with applications to biology and neuroscience. Several articles published in top international peer-reviewed venues, including ECML PKDD, SIAM Alenex, COSYNE, CCN and AISTATS.
I teach in the following master's level courses at KTH: - Machine Learning (Pr. Atsuto Maki) - Artificial Intelligence (Pr. Jana Tumova, Dr. André Pereira and Pr. Iolanda Leite) - Deep Learning in Data Science (Pr. Josephine Sullivan) It consists in teaching to master students, correcting and grading their assignments, presentations, projects and exams.
Supervised the master's thesis of Adrien Jouanny from KTH side, it was a collaboration between the Swiss Data Science Center (SDSC), the Paul Scherrer Institute (PSI) and the Swiss Tropical and Public Health Institute (TPH). His work supervisors were Dr. Imad El Haddad, Dr. Daniel Trejo Banos and Dr. Ekaterina Krymova. Thesis title: Leveraging Machine Learning methods alongside chemical transport, weather and land use data for organic aerosols component estimation
I help decision-makers solve concrete problems in companies and institutions by choosing the right mathematical and data-driven approach — AI-based or not — to reduce cost and time to impact. "Aniss is very talented and immediately understood and gave us advice on how to solve an AI/math problem for a system we are developing." — Mattias Nyman, CEO at Flygresor.se, Sweden’s largest flight comparison site.
I teach the core course of the Data Science program at DIS. Course description: In our increasingly digitized society, with sensors embedded in our bodies, equipment, and surroundings, we are generating, collecting, and storing data at unprecedented rates. Within this vast sea of data lie insights crucial for understanding, predicting, and impacting every aspect of our existence, including human behavior, financial trends, sustainable development, and health and illness. Extracting these insights requires careful execution at each step in the data analytics pipeline. In this course, we will take a hands-on approach to explore the key steps in the data analytics pipeline: data gathering, curation, and transformation; the use of computational and statistical tools to analyze both small and large datasets; and data visualization and reporting of analytical insights. Through real-world case scenarios, we will also evaluate and reflect on the validity of the analytical models.
DIS: DIS is a non-profit study abroad foundation established in Denmark in 1959 that offers semester, academic year, and summer programs taught in English in Copenhagen and Stockholm. We are designed for upper-division undergraduate students from top North American colleges and universities, providing a rigorous, cutting-edge curriculum enriched by experiential learning and faculty-led Study Tours across Europe. Course: Machine learning utilizes data to develop mathematical models capable of solving problems we have not been able to solve before. This course offers a hands-on approach to the theory and practice of machine learning, with real-world applications. It focuses on understanding various types of data, how to preprocess that data to make it useful for machine learning, approaches on how to solve different problems using machine learning, and how to improve a machine learning system.
Working on machine learning and ElectroEncephaloGraphy-based Brain Computer-Interface at the Robotics Perception and Learning division. The RPL division is part of the Department of Intelligent Systems at the School of Electrical Engineering and Computer Science of KTH. Topics: - meta imitation learning applied to robotics simulation. How to teach to a robot to imitate another robot? - EEG-based non-invasive BCI for analysing brain responses to visual stimuli, RSVP-based BCI (Rapid Serial Visual Presentation). How to communicate by thoughts with a computer which shows us images? Supervised by Dr. Ali Ghadirzadeh (postdoctoral researcher at Stanford University) and Prof. Danica Kragic Jensfelt (RPL/KTH).
Master thesis in Machine Learning and Topological Data Analysis at the Robotics Perception and Learning (RPL) division of the Royal Institute of Technology KTH Title: Towards topology-aware Variational Auto-Encoders: from InvMap-VAE to Witness Simplicial VAE Link to the thesis: http://urn.kb.se/resolve?urn=urn%3Anbn%3Ase%3Akth%3Adiva-309487 Background: Topological Data Analysis (TDA) is a recent field in data science aiming to study the "shape" of the data, or in other words to understand, analyse and exploit the toplogical and geometrical structure of the data, in order to get relevant information. For that purpose, it combines mathematical notions essentially from algebraic topology, geometry, combinatorics, probability and statistics, with powerful tools and algorithms studied in computational topology. Algebraic topology allows to identify objects that we can deform continuously (without breaking) from one to the other, and computational topology studies the application of computation to topology by developing algorithms aiming to construct and analyse topological structures. A widely used family of algorithms nowadays in artificial intelligence are the artificial neural networks because they succeed very well in many applications. In this master thesis we will investigate the use of TDA in order to better understand some artificial neural networks models called Variational Auto-Encoders, and to modify these algorithms with the hope that it will lead to an improvement of their performances. Goal: Aiming to improve the latent space representations in Variational Auto-Encoders by enforcing the topological structure of the input space to be preserved in the latent space. Our hypothesis is that this will allow a more efficient interpolation in the latent space, which can be beneficial for multiple applications like motion planning in robotics. Main supervisor: Dr. Anastasiia Varava & Co-supervisor: Vladislav Polianskii Examinator: Prof. Danica Kragic Jensfelt