PhD Position F/M Theory for Discrete Diffusion Models

inria-alumni; Paris; Hybrid; EUR 25,000 - 35,000; Full time

Benefits offered by this job

Partial reimbursement of public travel

Leave: 7 weeks + 10 RTT days + special

Teleworking and flexible hours

Professional equipment provided

Social, cultural and sports events

Job summary

Inria recrute un doctorant pour le cadre ERC Starting Grant DYNASTY sur la théorie des modèles de diffusion discrets. Le/la doctorant(e) étudiera les fondements théoriques et validera les résultats par des expériences numériques, avec publication des travaux dans des revues et conférences internationales.

Le poste exige un haut niveau d'anglais et des compétences en Python et bibliothèques DL, avec possibilité de déplacements à des conférences.

Qualifications

• Anglais professionnel à un niveau élevé requis.
• Compétences solides en Python et bibliothèques de DL.
• Connaissances en théorie des modèles de diffusion ou apprentissage automatique souhaitées.

Responsibilities

• Mener des recherches théoriques sur les modèles de diffusion Discrèts.
• Réaliser des expériences numériques pour valider les résultats.
• Rédiger des articles scientifiques et présenter les travaux dans des conférences.
• Diffuser les résultats scientifiques dans des forums appropriés.

Skills

Anglais professionnel

Python

Education

PhD en informatique / ML

Tools

Deep learning libraries

Job description

PhD Position F/M Theory for Discrete Diffusion Models

Fonction : Doctorant

The PhD position will be in the framework of the ERC Starting Grant DYNASTY (Dynamics-Aware Theory of Deep Learning).

The position might include traveling to conferences for paper presentation. Travel expenses will be covered within the limits of the scale in force.

Discrete diffusion models have become a leading alternative to autoregressive models for generating text and other categorical data [1, 2]. They generate samples by reversing a Markov chain that progressively corrupts the data, using a score function learned by a neural network [1]. Despite their empirical success, their theoretical understanding remains far less developed than that of their continuous counterparts.

The objective of this PhD is to study the theoretical foundations of discrete diffusion models, and in particular how well they learn and generalize when the score is estimated by a neural network. Building on recent statistical analyses [3, 4] and on generalization bounds for diffusion models [5], the project will investigate how statistical, approximation and optimization errors affect the quality of the generated samples, and will validate the theory through numerical experiments.

References
[1] A. Lou, C. Meng, S. Ermon. Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution. ICML, 2024.
[2] S. S. Sahoo, M. Arriola, Y. Schiff, A. Gokaslan, E. Marroquin, J. T. Chiu, A. Rush, V. Kuleshov. Simple and Effective Masked Diffusion Language Models. NeurIPS, 2024.
[3] Z. Zhang, H. Fu, Z. Yang, M. Wang, T. Zhao, M. Chen. Generalization Bounds for Discrete Diffusion: Statistical Advantage of Masking. ICML, 2026.
[4] L. Blunda, T. Farghly, U. Şimşekli, A. Durmus. Discrete Diffusion Models are Minimax Optimal Under Sparsity. arXiv:2610.19020, 2026.
[5] B. Dupuis, D. Shariatian, M. Haddouche, A. Durmus, U. Şimşekli. Algorithm- and Data-Dependent Generalization Bounds for Diffusion Models. arXiv:2506.03849, 2026.

Conduct theoretical research

Conduct experiments for empirical verification

Write scientific articlesDisseminate the scientific work in appropriate venues.

Technical skills and level required :

Languages : High-level of professional/academic English

Coding skills : Good level of coding in Python and related deep learning libraries

Avantages

• Partial reimbursement of public transport costs
• Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
• Possibility of teleworking and flexible organization of working hours
• Professional equipment available (videoconferencing, loan of computer equipment, etc.)
• Social, cultural and sports events and activities