The GHOST team had several papers accepted at NeurIPS 2025!

  1. Does Stochastic Gradient really succeed for bandits? accepted as oral!
    • About: Regret guarantees of Stochastic Gradient Bandit in reinforcement learning, a policy gradient approach with softmax parametrization.
  2. Non-stationary Bandit Convex Optimization: A Comprehensive Study.
    • About: New algorithms for non-stationary bandit convex optimization that achieve state-of-the-art regret with respect to standard non-stationarity measures.
  3. Efficient Quadratic Corrections for Frank-Wolfe Algorithms.
    • About: Frank-Wolfe algorithms with corrective steps for convex quadratic objectives, generalizing existing variants and improving efficiency.
  4. Efficient Kernelized Learning in Polyhedral Games Beyond Full-Information: From Colonel Blotto to Congestion Games.
    • About: Efficient learning of coarse correlated equilibria in polyhedral games with exponentially large action spaces under partial information (semi-bandit, bandit,…).
  5. Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration as spotlight!
    • About: The long-run behavior of multi-agent regularized learning in continuous games under stochastic uncertainty.
  6. Robust Equilibria in Continuous Games: From Strategic to Dynamic Robustness.
    • About: Introducing a notion of robust equilibrium for continuous games (equilibria that remain invariant under payoff perturbations), and a connection between strategic and dynamic robustness under regularized learning.