Younesse Kaddar

Younesse Kaddar

Hi! My name is Younesse Kaddar. I'm a computer science PhD student at the University of Oxford, UK. Welcome to my website!

About Me

Hi, I’m Younesse Kaddar, a French PhD student in Computer Science at the University of Oxford working on category theory, probabilistic programming, and deep learning for reasoning, especially program synthesis and automated theorem proving. Ultimately, what I’m passionate about is understanding cognition, and I see two paths toward this goal:

  • A more algebraic one through formal methods and category theory (deeper mathematical understanding, but less immediately applicable); a first-step sweet spot on this front (in my very biased opinion) is the categorical semantics of probabilistic programming languages, and, more recently, Markov categories.
  • And a more empirical one through state-of-the-art transformer-based reasoning (which has shown incredibly surprising results lately, but still lacks reliability, steerability, safety, and formal theory).

Last year, I did research internships to explore the latter direction. I spent six months at Mila (Quebec AI Institute) working in Yoshua Bengio’s team on GFlowNets for steering outputs of large language models (LLMs), trying to make LLMs more reliable in their reasoning with amortized sampling of high-quality reasoning chain-of-thoughts. I then joined Cohere For AI as a research scholar, focusing on detecting and mitigating LLM hallucinations.

Currently, I’m back in my PhD, working on LLM-guided probabilistic program synthesis: the idea is to use LLMs to automatically turn natural language descriptions into probability distributions over statistical models (expressed as probabilistic programs) to model real-world phenomena. The dream would be to make complex statistical modeling more accessible and interpretable while ensuring safety, by using GFlowNet-finetuned LLMs as “world model compilers” rather than unrestricted agents.

I am also a co-maintainer of LazyPPL, a Haskell probabilistic programming library for Bayesian nonparametrics, within Sam Staton’s team.

So more broadly, my academic work tries to bridge two approaches: formal methods/category theory on one hand, practical machine learning (especially LLM research) on the other, with the long-term goal of building AI systems we can trust, to, ultimately, better understand the nature of intelligence.

Outside of academia, I’m also the cofounder and CTO of RightPick, a startup which helps alumni from a network of European universities find top jobs in tech, finance, and consulting. Our iOS app is available on the AppStore, and RightPick has been featured on several sections of the Oxford University Careers website, including the Management Consultancy, Technology, Data, Machine Learning & AI, Business & Management, and Banking & Investment pages.

Publications and Research Experience

2025

  • Uncertainty-Aware Step-wise Verification with Generative Reward Models
    Z. Ye, L. C. Melo, Y. Kaddar, P. Blunsom, S. Staton, Y. Gal
    Preprint 2025
    Paper

2024

  • Can a Bayesian Oracle Prevent Harm from an Agent?
    Y. Bengio, M. K. Cohen, N. Malkin, M. MacDermott, D. Fornasiere, P. Greiner, Y. Kaddar
    Paper Code

  • Amortizing Intractable Inference in Large Language Models
    E. J. Hu, M. Jain, E. Elmoznino, Y. Kaddar, G. Lajoie, Y. Bengio, N. Malkin
    International Conference on Learning Representations (ICLR) 2024 (Honourable Mention)
    Paper Code

  • Probabilistic Programming Interfaces for Random Graphs: Markov Categories, Graphons, and Nominal Sets
    N. Ackerman, C. Freer, Y. Kaddar, J. Karwowski, S. Moss, D. Roy, S. Staton, H. Yang
    Principles of Programming Languages (POPL) 2024
    Paper

2023

  • A Model of Stochastic Memoization and Name Generation in Probabilistic Programming: Categorical Semantics via Monads on Presheaf Categories
    Y. Kaddar and S. Staton
    Mathematical Foundations of Programming Semantics (MFPS) 2023
    Paper Slides

  • Affine Monads and Lazy Structures for Bayesian Programming
    S. Dash, Y. Kaddar, H. Paquet and S. Staton
    Principles of Programming Languages (POPL) 2023
    Paper Website Code

Conference Presentations

  • HOPE 2022: Higher order programming with probabilistic effects: A model of stochastic memoization and name generation
    Younesse Kaddar and Sam Staton
    Link Video

  • Applied Category Theory (ACT) 2022 (Aug. 2022): Statistical Programming with Categorical Measure Theory and LazyPPL (demo)
    S. Dash, Y. Kaddar, H. Paquet and S. Staton
    Slides Video

Conference Panels

  • Panelist at PADL 2024 (POPL workshop)
    Topic: Declarative Languages for Safe AI
    Chair: Ekaterina Komendantskaya (Heriot-Watt University & University of Southampton)
    Website Video

Projects

RightPick (2023-present)

Cofounder & CTO
Website App Store

Open Source Projects

LazyPPL (2021-present)

Co-maintainer of this Haskell-based probabilistic programming library for Bayesian nonparametrics (Sam Staton’s team)
Website GitHub

Awards

2023

  • 1st Prize & Impact Prize, Bio x ML Hackathon 2023, HuggingFace, OpenBioML, Lux Capital & LatchBio
    Project: SVM - Generate unified protein embedding across multiple protein modalities
    Code 🤗 Model

2022

Education

2020-present

DPhil in Computer Science, University of Oxford, UK
Scholarship: Oxford-DeepMind
Teaching and tutoring: Principles of Programming Languages, Bayesian Statistical Probabilistic Programming

2019-2020

Visiting researcher (Predoctoral research year), University of Cambridge, UK
Department of Computer Science and Technology
Supervisor: Marcelo Fiore

2018-2019

Parisian Master of Research in Computer Science (MPRI): 2nd year (M2R; Masters 2 Research)
École Normale Supérieure Paris-Saclay, Paris
Honors: Summa cum laude

2017-2018

Parisian Master of Research in Computer Science (MPRI): 1st year (M1; Masters 1)
École Normale Supérieure Paris-Saclay, Cachan / Paris
Overall rank: 1st (out of 27)
Courses: Category theory & λ-calculus, Advanced Complexity, Statistical Learning, Computer Vision, Robot Motion Planning, Initiation to Research, English
Extra courses: Proof assistants (LMFI Master, Paris-Diderot), Modules and finite groups (Math Master at École Polytechnique)

Cogmaster (Cognitive Science)
École Normale Supérieure Paris
Courses: Computational Neuroscience, Neuromodeling, Neurorobotics, Machine Learning applied to Neuroscience

2016-2017

Bachelor of Computer Science
École Normale Supérieure Paris-Saclay, Cachan
Overall rank: 1st (out of 28)
Courses: λ-calculus & Logic, Logic Projects (DPLL algorithm & Coq project), Discrete Mathematics, Programming & Semantics, Advanced Programming, Compiler Project, Formal Languages, Computability & Complexity, Algorithmics, Advanced Algorithms, Abstract Algebra, English, Computer Architecture

2013-2016

Classes Préparatoires aux Grandes Écoles, MPSI-MP*
Lycée Henri Poincaré, Nancy
Preparatory courses to nationwide competitive exams in mathematics, physics and computer science

2012-2013

Baccalauréat S, Lycée Henri Poincaré, Nancy
Major in mathematics, with highest honors

Research Internships

  • Cohere For AI Scholars Programme (Jan. 2024 – Aug. 2024)
    Research Topic: LLM Hallucinations
    Mentor: Beyza Ermiş

  • PhD Internship at Mila (June 2023 – Jan. 2024), Quebec Artificial Intelligence Institute, Université de Montréal
    Supervisor: Yoshua Bengio
    Research Topic: GFlowNets for reasoning & AI safety

  • Pre-doctoral Internship (Oct. 2019 – Aug. 2020), University of Cambridge, Department of Computer Science
    Title: Ideal Distributors
    Supervisor: Marcelo Fiore
    Cambridge Internship Report

  • M2 Internship (Apr-Aug 2019), Macquarie University
    Title: Tricocycloids, Effect Monoids and Effectuses
    Supervisor: Richard Garner
    Sydney Internship Report

  • M1 Internship (June – Aug 2018), University of Oxford
    Title: Event Structures as Presheaves
    Supervisor: Ohad Kammar
    Oxford Internship Report

  • L3 Internship (June – Aug 2017), University of Nottingham
    Title: Type Theory forms a weak omega groupoid
    Supervisors: Thorsten Altenkirch, Paolo Capriotti, Nicolai Kraus
    Nottingham Internship Report

Teaching Experience

University of Oxford, Department of Philosophy, 2023

  • Topics in Minds and Machines: Perception, Cognition, and ChatGPT
    Philosophy Seminar, University of Oxford
    Role: Lectured on Deep Learning and Large Language Models

University of Oxford, Department of Computer Science, 2022

  • Bayesian Statistical Probabilistic Programming, University of Oxford
    Role: Class tutor and marker
  • Imperative Programming in Scala III, University of Oxford
    Role: Demonstrator

University of Oxford, Department of Computer Science, 2020-2021

  • Principles of Programming Languages, University of Oxford
    Role: Class tutor and marker, personal tutor (Exeter College)

Tutorial Teaching at ENS Paris-Saclay (2017-2018)

  • Subjects: Computability and Complexity Theory, Algorithms, Automata Theory, Formal Language Theory
    Role: Personal tutor

Online Research Programme at Immerse Education (Jul. 2020, Dec. 2020, Jul. 2021)

  • Topics: Bayesian probabilistic programming, Algorithms, Supervised learning
    Role: Personal tutor and supervisor

Additional Training & Professional Development

Online Courses and Reading Groups

  • AI Alignment Course (12 weeks), BlueDot Impact Comprehensive technical AI safety curriculum focused on reducing risks from advanced AI systems
  • AI Governance Reading Group, University of Oxford

Summer Schools

  • Summer School in Neurosymbolic Programming (June 2024) Salem, Massachusetts, USA
  • MIT Probabilistic Programming Mini School (August 2023)
  • CalTech Neurosymbolic Programming Summer School (July 2022)
  • Oregon Programming Languages Summer School (June-July 2022)

Languages

  • French: First language
  • English: Fluent (Cambridge Certificate in Advanced English, Score: 201/210, CEFR Level: C2)
  • German: Basic (CEFR Level: B2)