We are proud to announce that our very first Optimization and Learning Day will take place in Grenoble on June 17th, 2026 at the IMAG building on the main university campus, see location here.
The building can be accessed from the train station with the B tram line.

This workshop will be a small-scale gathering with around 50 attendees, dedicated to the latest advances that combine decision-making (continuous or discrete optimization, operations research) with machine learning.

The workshop will follow a single-track format with invited speakers. We welcome submissions for the poster session to foster interactions between participants.

Practical details

Lunch options

Lunch is self-organised, there are many options on campus. CROUS operates several student cafeterias around campus. Most offer to-go options. You’ll find a map of locations here.

Otherwise, here are the nearest options (these are not endorsements). It’s preferable to book ahead at restaurants.

Takeaway

  • La Tab’ Verte (1103 Rue des Universités, 38610 Gières)
  • Little India (1103 Rue des Universités, 38610 Gières)
  • EVE - Espace Vie Etudiante, in front of the workshop building (101 Pl. du Torrent, 38400 Saint-Martin-d’Hères)

Restaurants

  • L’Oiseau Blanc (800 Av. Centrale, 38400 Saint-Martin-d’Hères), reservation required: oiseau.blanc@crous-grenoble.fr
  • Vertical’Art Grenoble (41 Rue des Glairons, 38400 Saint-Martin-d’Hères, 09 81 71 49 01)
  • O’Campus (430 Rue de la Passerelle, 38400 Saint-Martin-d’Hères, 09 81 51 45 40)
  • Le Martin’s cafe (401 Av. de la Bibliothèque, 38400 Saint-Martin-d’Hères, 04 76 54 43 36)

Program

Mathieu and Guillaume – 9.00 - 9.10 – Welcome

Silvia di Gregorio – 9.10 - 9.50

Partial Optimality in Cubic Correlation Clustering for General Graphs Abstract
The higher-order correlation clustering problem for a graph G and costs associated with cliques of G consists in finding a clustering of G so as to minimize the sum of the costs of those cliques whose nodes all belong to the same cluster. To tackle this NP-hard problem in practice, local search heuristics have been proposed and studied in the context of applications. Here, we establish partial optimality conditions for cubic correlation clustering, i.e., for the special case of at most 3-cliques. We define and implement algorithms for deciding these conditions and examine their effectiveness numerically, on two data sets.

Michael Arbel – 9.50 - 10.30

PEIRA: Learning Predictive Encoders through Inter-View Regressor Alignment Abstract
Non-contrastive self-supervised learning (SSL) is an effective framework for predictive representation learning, but popular (and in practice effective) methods such as SimSiam, BYOL, I-JEPA or DINO, which rely on a form of self-distillation to train a teacher-student network, remain poorly understood as they typically do not minimize a well-defined objective. We analyze the dynamics of a variant of the Joint Embedding Predictive Architecture (JEPA) using a regularized linear regressor to predict the learned representations of two views of the data from one another, and fully characterize its stability: non-collapsed stable equilibria align with leading nonlinear canonical correlation subspaces, while collapsed equilibria may also be stable attractors. Motivated by this result, we introduce PEIRA, a non-contrastive SSL method with an explicit objective defined through the trace of the optimal linear regressor. We show that its only stable equilibria are nontrivial global minimizers and recover the same canonical correlation subspaces, with regularization selecting the effective dimension. Experiments on ImageNet-1K and CIFAR-10 show PEIRA is competitive with VICReg and LeJEPA baselines, and qualitative empirical results support the theory.

Coffee break – 10.30 - 11.00

Francesca Demelas – 11.00 - 11.40

Bundle Network: Unrolling the Bundle Method with Recurrent Attention Abstract
We introduce Bundle Network, a learning-based algorithm for convex non-smooth minimization inspired by the classical Bundle Method. While traditional approaches require heuristic tuning of a regularization parameter, our method learns to adapt it automatically from data. In addition, we replace the iterative quadratic subproblem that computes the search direction, classically solved as a convex combination of subgradients at previously visited points, with a recurrent neural network enhanced with an attention mechanism. The full algorithm is unrolled as a computation graph, enabling end-to-end training via automatic differentiation. We evaluate our approach on Lagrangian dual relaxations of two combinatorial optimization problems (Multi-Commodity Network Design and Generalized Assignment) and show that Bundle Network consistently outperforms grid-search baselines while generalizing well across problem instances.

Senne Berden – 11.40 - 12.20

Solver-Free Decision-Focused Learning for Linear Optimization Problems Abstract
Mathematical optimization is a fundamental tool for decision-making in a wide range of applications. However, in many real-world scenarios, the parameters of the optimization problem are not known a priori and must be predicted from contextual features. This gives rise to predict-then-optimize problems, where a machine learning model predicts problem parameters that are then used to make decisions via optimization. A growing body of work on decision-focused learning (DFL) addresses this setting by training models specifically to produce predictions that maximize downstream decision quality, rather than accuracy. While effective, DFL is computationally expensive, because it requires solving the optimization problem with the predicted parameters at each loss evaluation. In this talk, I will address this computational bottleneck for linear optimization problems, a common class of problems in both DFL literature and real-world applications. I will present a solver-free training method that exploits the geometric structure of linear optimization to enable efficient training with minimal degradation in solution quality. The method is based on the insight that a solution is optimal if and only if it achieves an objective value that is at least as good as that of its adjacent vertices on the feasible polytope. Building on this, the method compares the estimated quality of the ground-truth optimal solution with that of its precomputed adjacent vertices, and uses this as loss function. Experiments demonstrate that the method significantly reduces computational cost while maintaining high decision quality.

Lunch break – 12.20 - 13.40

Marianne Defresne – 13.40 - 14.20

Emmental-PLL: Efficient Neuro-Symbolic Learning of Constraints & Objective

Abstract
Real-life decision making often involves reasoning on ill-defined problems, where exact constraints or parameters (such as costs) are unknown. We aim to automatize the problem definition by learning it from natural input. We introduce a differentiable neuro-symbolic architecture and its dedicated loss, the Emmental-PLL*. The gap between combinatorial optimization and learning is bridged with discrete graphical models. On the popular benchmark of learning how to play Sudoku and its visual variant, our method is able to learn the rules from example grids. It requires a fraction of time and data needed by existing methods and stands out for its ability to scale to large problems. On a visual Min-Cut/Max-cut task, it optimizes the regret as well as a Decision-Focused-Learning regret-dedicated loss. Finally, it efficiently learns the energy optimisation formulation of the large real-world problem of designing proteins.

Defresne, M., Gambardella, R., Barbe, S., & Schiex, T. (2026). Scaling Neuro-symbolic Problem Solving: Solver-Free Learning of Constraints and Objectives. Journal of Artificial Intelligence Research, 85.

Sofia Michel – 14.20 - 15.00

Generalizable Neural Combinatorial Optimization — One Model to Solve Them All? Abstract
Recent years have seen considerable progress in machine learning–based heuristics for combinatorial optimization, particularly constructive neural heuristics that use neural networks to build solutions step by step. Despite their success across a variety of combinatorial tasks, these methods often require specialized models trained separately for each problem and instance distribution. In this talk, I will discuss recent work on improving the generalization of neural combinatorial optimization approaches. I will introduce a framework that leverages the symmetries inherent in many combinatorial problems to boost the out-of-distribution generalization. Building on this framework, I will present GOAL, a generalist model capable of efficiently solving multiple combinatorial problems and which can be fine-tuned to handle new ones. By tackling a broad range of tasks — including routing, scheduling, packing, location, and graph problems — GOAL represents a promising step toward developing a foundation model for combinatorial optimization.

Andrii Kliachkin & Gilles Bareilles – 15.00 - 15.40

humancompatible.train: An Open-Source Toolkit for Constrained Machine Learning Abstract
There has been a considerable interest in constrained training of deep neural networks (DNNs) recently, driven by the realization that constrained machine learning enables fairness-aware training, accurate implementations of physics-informed neural networks, and integration of symbolic domain knowledge into statistical learning. Several toolkits have been proposed for this task, yet there is still no industry standard. We present humancompatible.train (https://github.com/humancompatible/train), an easily-extendable PyTorch-based Python package for training DNNs with stochastic constraints. We implement multiple previously unimplemented algorithms for stochastically constrained stochastic optimization, including the new Stochastic Penalty-Barrier Method (https://arxiv.org/abs/2605.18618) with competitive performance on a fairness-constrained learning benchmark (https://arxiv.org/abs/2507.04033, ICLR 2026).

End of the workshop – 15.40. Participants can continue the conversation at the EVE center in front of the workshop building, which serves food and refreshments.

Acknowledgements

We gratefully acknowledge support from the following institutions: