Bias In, Bias Out? Finding Unbiased Subnetworks in Vanilla Models
I. Matos, A. Saoud, E. Iakovleva, V. Pastore and E. Tartaglione. "Bias In, Bias Out? Finding Unbiased Subnetworks in Vanilla Models." In CVPR'26.
CV
Work experience
- Postdoctoral Researcher, Nov 2023 - Apr 2026
- Multimedia team, Télécom Paris
- Research areas: Fairness, Model Debiasing, Model Pruning, Data Pruning, Machine Unlearning, Generative AI, Diffusion Models, Image Editing
- Duties included: co-supervison of Master and PhD students, teaching
- Supervisors: Enzo Tartaglione(2024-2026) and Stéphane Lathuilière (2023-2024)
- PhD Student / Reasearch Assistant, Oct 2018 - Dec 2022
- THOTH team, Inria Grenoble
- Research areas: Meta-Learning, Few-Shot Learning, Transfer Learning, Multi-Domain Learning, Hypernetworks
- Supervisors: Jakob Verbeek and Karteek Alahari
- Research Intern, Feb 2018 - Sep 2018
- THOTH team, Inria Grenoble
- Research areas: Generative Models, Autoregressive Models, PixelCNN
- Supervisor: Jakob Verbeek
- Research Intern, Jun 2016 - Aug 2016
- Sberbank Technology, joint project with Bayesian Methods Research Group
- Project: prediction of customer’s needs by clustering customers using Latent Dirichlet Allocation (LDA) on transactions
- Duties included: cleaning and pre-processing of data, training of models, analysis and interpretation of the results, communication between the academic research team and the team at Sberbank Technology, task assignment and coordination, preparation and presentation of the final report
- Research Intern, Dec 2015 - Jun 2017
- Bayesian Methods Research Group
- Research areas: Bayesian Methods, Deep Learning, Computer Vision
- Supervisor: Dmitry Vetrov
Education
- Ph.D in Mathematics and Informatics, Université Grenoble Alpes, 2022
- M.S. in Industrial and Applied Mathematics (“with distinction”), Institut Polytechnique de Grenoble, 2018
- M.S. in Computer Science (“with distinction”), Skolkovo Institue of Science and Technology, 2017
- B.S. in Applied Mathematics and Physics, Moscow Institue of Physics and Technology, 2015
Skills
- Research Areas: Fairness, Trustworthy AI, Debiasing, Model Pruning, Efficient AI, Machine Unlearning, Generative AI, Diffusion Models, Transfer Learning, Meta-Learning, Few-Shot Learning, Bayesian Methods
- Technical Experience: Python, PyTorch, Git
- Previous Experience: TensorFlow, R, SQL
- Languages: English (fluent), French (intermediate), Russian (native)
Publications
How I Met Your Bias: Investigating Bias Amplification in Diffusion Models
N. Roos, E. Iakovleva, A. Gjergji, V. Pastore and E. Tartaglione. "How I Met Your Bias: Investigating Bias Amplification in Diffusion Models." In WACV'26.
Specify and Edit: Overcoming Ambiguity in Text-Based Image Editing
E. Iakovleva, F. Pizzati, P. Torr and S. Lathuilière. "Specify and Edit: Overcoming Ambiguity in Text-Based Image Editing." In BMVC'25.
Multi-Domain Learning with Modulation Adapters
E. Iakovleva, K. Alahari and J. Verbeek. "Multi-Domain Learning with Modulation Adapters." In arXiv:2307.08528'23.
Meta-Learning with Shared Amortized Variational Inference
E. Iakovleva, J. Verbeek and K. Alahari. "Meta-Learning with Shared Amortized Variational Inference." In ICML'20.
Exploring Generative Image Modeling with Block PixelCNNs
E. Iakovleva. "Exploring Generative Image Modeling with Block PixelCNNs." MSc Thesis at Grenoble INP, 2018.
Multimodal Distributions in Variational Autoencoders
E. Iakovleva. "Multimodal Distributions in Variational Autoencoders." MSc Thesis at Skoltech, 2017.
Community Service
- Reviewer for journals: IEEE ToM, CVIU, TMLR
- Reviewer for conferences: ICML, ICCV, ECCV, CVPR, WACV, BMVC
- Outstanding reviewer: ICML 2022 (top 10%)
Talks
Text-Based Image Editing at BMVC’25
Online presentation at British Machine Vision Conference, Sheffield, UK
Meta-Learning at ICML’20
Online presentation at International Conference on Machine Learning, Vienna, Austria (Virtual Only)