Paper Accepted at NeurIPS 2026

Our paper “Dynamic Regret in Online Convex Optimization with Indicator Switching Costs,” with George Iosifidis, has been accepted at NeurIPS 2026.

The paper studies online decisions that incur a fixed cost whenever they change, capturing startup overheads such as server activation, model deployment, and cache updates. Existing guarantees for such costs only compete against a fixed benchmark. We propose a meta-learning algorithm whose dynamic regret is minimax-optimal, up to logarithmic factors, for tracking piecewise-constant comparators, without prior knowledge of how often the comparator changes.

• Preprint: arXiv:2609.30556 📄