Institution profile

Institut Polytechnique de Paris

Academic institutioneurope · fr
Official website
Research library838linked papers
Opportunities0open roles
Selected work

Representative Papers

Plug-and-Play image restoration with Stochastic deNOising REgularization

Feb 01, 2024International Conference on Machine Learning

Plug-and-Play (PnP) algorithms apply denoisers to progressively noise-decaying iterates, conflicting with diffusion models (DMs), which deploy denoisers exclusively on controllably noisy data. This inconsistency undermines theoretical alignment and practical performance. Method: We propose SNORE—a stochastic noise-level-adaptive regularization framework for image inverse problems (e.g., deblurring, inpainting). SNORE constructs a noise-level-matched stochastic gradient descent optimizer via explicit noise-aware regularization. Contribution/Results: This is the first PnP method to incorporate noise-perceptive stochastic regularization, unifying PnP and DM denoising logic while providing rigorous convergence and annealing-theoretic analysis. Experiments demonstrate that SNORE, when integrated with deep denoisers (e.g., DnCNN), achieves state-of-the-art performance on deblurring and inpainting—outperforming prior methods in PSNR, SSIM, and perceptual quality.

8 citations2 influentialRead paper

Generalization Bounds of Surrogate Policies for Combinatorial Optimization Problems

Jul 24, 2024arXiv.org

In combinatorial optimization, the empirical risk w.r.t. model parameters is piecewise constant, hindering gradient-based optimization and lacking theoretical generalization guarantees. Method: For contextual stochastic optimization with complex objectives, we propose a perturbation-driven risk smoothing strategy. Our approach integrates statistical learning models with a surrogate combinatorial optimization oracle to construct a context-aware, generalization-controllable decision framework. Contribution/Results: We establish the first unified generalization bound incorporating perturbation bias, statistical error, and optimization error. We introduce the notion of “uniform weak consistency” to characterize the coupled stability between the learning model and the surrogate oracle, proving its universality under mild assumptions. Experiments on stochastic vehicle scheduling demonstrate strong generalization performance. This work provides the first verifiable theoretical generalization framework for contextual stochastic optimization.

5 citationsRead paper

Efficient Hamiltonian, structure and trace distance learning of Gaussian states

Nov 05, 2024arXiv.org

This work addresses the efficient learning of thermal bosonic Gaussian states at positive temperature, aiming to jointly infer the underlying quadratic Hamiltonian parameters, interaction graph structure, and approximate the state in trace distance—using only heterodyne measurements and with low sample complexity. Under bounded temperature, squeezing magnitude, displacement amplitude, and graph maximum degree, we propose a method based on local covariance submatrix estimation and a novel local inversion technique, circumventing full covariance matrix estimation. Leveraging continuity bounds between Hamiltonians and covariance matrices, we achieve, for the first time, quadratic-precision learning of Gaussian states in trace distance: achieving error ε requires only $O(varepsilon^{-2})$ samples, with polynomial dependence on system parameters. The algorithm is both sample- and computationally efficient, providing the first practical framework for continuous-variable quantum state learning with rigorous trace-distance guarantees.

4 citationsRead paper

Contrastive Knowledge Distillation for Embedding Refinement in Personalized Speech Enhancement

Apr 06, 2025IEEE International Conference on Acoustics, Speech, and Signal Processing

This work addresses the limitations of conventional personalized speech enhancement methods, which rely on pre-extracted static speaker embeddings that struggle to adapt to target speaker variations during inference and require computationally expensive upstream models for high-quality embeddings. To overcome these challenges, the authors propose a lightweight speaker encoder with only 150K parameters, coupled with a contrastive knowledge distillation strategy tailored for embedding optimization. This approach enables dynamic refinement of speaker representations at inference time by efficiently distilling discriminative features from a complex teacher model. The proposed method achieves significant performance gains in speech enhancement while maintaining minimal computational overhead.

3 citationsRead paper

Contextual Causal Bayesian Optimisation

Jan 29, 2023arXiv.org

Existing causal optimization methods often neglect observational variables, leading to suboptimal interventions. Method: This paper proposes Context-aware Causal Bayesian Optimization (CaCBO), a novel framework that explicitly models observational variables as contextual information to guide controllable interventions; introduces a multi-armed bandit–driven strategy range selection mechanism to overcome the failure of conventional causal acquisition functions under context-dependent settings; and constructs a context-adaptive acquisition function. Contribution/Results: We provide theoretical guarantees showing that CaCBO’s cumulative regret strictly improves upon both standard Contextual Bayesian Optimization (CoBO) and Causal Bayesian Optimization (CaBO). Empirically, CaCBO achieves sublinear regret convergence across diverse causal environments, significantly enhancing both the efficiency and robustness of discovering optimal intervention policies.

2 citations1 influentialRead paper
Recent publications

Latest Papers