Id: arxiv/2607.03232 Url: https://arxiv.org/abs/2607.03232 Tags: math.CO, stat.ML Score: 0 Title: Fast SDP certification of neural networks : towards large multi-class datasets Domain: math.CO Source: arxiv Authors: Margot Boyer, Clément Rambour, Zacharie Alès, Amélie Lambert Reviews: 0 Subject: math.CO Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-03 Page: https://delphi.science/r/arxiv/2607.03232 We present a new quadratic model for the certification problem in adversarial robustness, which simultaneously accounts for all possible target classes. Building on this model, we propose a novel semidefinite programming (SDP) relaxation for incomplete verification. A key advantage of our approach is that it certifies robustness in a single optimization, avoiding the need for a separate resolution per class. This yields a significant computational speed-up and enables scalability to large datasets with many classes. To further improve efficiency, we also propose an effective pruning strategy…