A selection of recent talks. Click "slides" to download the deck.
A selection of recent talks. Click "slides" to download the deck.
SIAM Optimization
Edinburgh, UK · June 2026
This talk focuses on large-scale bilevel optimization problems that exhibit a natural block-separable structure, a setting that frequently appears in modern machine learning pipelines, large data-cleaning tasks, and parametric model selection. Such problems pose substantial computational challenges due to the interaction between high-dimensional upper-level variables and the need to repeatedly solve lower-level optimization problems. To address these difficulties, we propose a randomized block-proximal framework that updates only a small subset of hyperparameters at each iteration, thereby reducing both memory usage and computational cost. The method relies on implicit differentiation to compute hypergradients, but does so in a scalable, matrix-free manner that avoids forming or storing large Hessian matrices. This combination yields an approach that is well-suited for problems with large number of parameters and offers a practical alternative to traditional full-batch bilevel solvers.
slidesApplied Inverse Problems
Rio de Janeiro, Brazil · September 2023
We study a bilevel optimization framework for hyperparameter tuning of variational models, with a focus on sparse regression and classification tasks. In particular, we consider a weighted elastic-net regularizer, where feature-wise regularization parameters are learned through a bilevel formulation. A key novelty of our approach is the use of the Forward-Backward Envelope (FBE) to smooth the nonsmooth lower-level problem while preserving its set of minimizers. This reformulation yields a bilevel objective composed with a locally Lipschitz solution map, allowing the application of generalized subdifferential calculus to derive basic subgradients and enable efficient subgradient-based optimization. Empirical results on synthetic datasets demonstrate that our approach significantly outperforms scalar regularization methods in terms of prediction accuracy and support recovery. These findings highlight the benefits of feature-wise regularization and the effectiveness of bilevel optimization as a principled framework for learning interpretable and high-performing models.
slides