Publications

Also on Google Scholar.

2026

• Preprint
  Goldilocks RL: Tuning Task Difficulty to Escape Sparse Rewards for Reasoning

  Ilia Mahrooghi, Aryo Lotfi, Emmanuel Abbe

  A learned teacher picks prompts near a 50% pass rate for GRPO training, avoiding zero-variance gradients and cutting compute by up to 78%.

  arXiv PDF

2025

• UAI 2025
  Multi-armed Bandits with Missing Outcome

  Ilia Mahrooghi, Mahshad Moradi, Sina Akbari, Negar Kiyavash

  Conference on Uncertainty in Artificial Intelligence (UAI 2025)

  Bandit algorithms for outcomes missing under MCAR, MAR, and MNAR mechanisms, with regret bounds for UCB-based methods.

  arXiv PDF Conference