From Data to Rewards: a Bilevel Optimization Perspective on Maximum Likelihood Estimation
Abstract
A bilevel optimization framework is used to align generative models with high-quality datasets in the absence of explicit reward signals, with applications in classification and model-based reinforcement learning.
Generative models form the backbone of modern machine learning, underpinning state-of-the-art systems in text, vision, and multimodal applications. While Maximum Likelihood Estimation has traditionally served as the dominant training paradigm, recent work have highlighted its limitations, particularly in generalization and susceptibility to catastrophic forgetting compared to Reinforcement Learning techniques, such as Policy Gradient methods. However, these approaches depend on explicit reward signals, which are often unavailable in practice, leaving open the fundamental problem of how to align generative models when only high-quality datasets are accessible. In this work, we address this challenge via a Bilevel Optimization framework, where the reward function is treated as the optimization variable of an outer-level problem, while a policy gradient objective defines the inner-level. We then conduct a theoretical analysis of this optimization problem in a tractable setting and extract insights that, as we demonstrate, generalize to applications such as tabular classification and model-based reinforcement learning. We release the code at https://github.com/abenechehab/nll_to_po .
Community
📢📢 New preprint and code alert!!
How can we leverage Policy Gradient methods (e.g. REINFORCE, GRPO) without having access to explicit reward signals?
💡 In our new work, “From Data to Rewards: a Bilevel Optimization Perspective on Maximum Likelihood Estimation”, we address this question by bridging the gap between Maximum Likelihood Estimation and Policy Gradient methods.
📜 Preprint: https://arxiv.org/abs/2510.07624
🖥️ Code: https://github.com/abenechehab/nll_to_po
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Beyond Imitation: Recovering Dense Rewards from Demonstrations (2025)
- Calibrating Generative Models (2025)
- Stabilizing Policy Gradients for Sample-Efficient Reinforcement Learning in LLM Reasoning (2025)
- Inference-Time Alignment Control for Diffusion Models with Reinforcement Learning Guidance (2025)
- On Discovering Algorithms for Adversarial Imitation Learning (2025)
- Expressive Value Learning for Scalable Offline Reinforcement Learning (2025)
- Distributional Inverse Reinforcement Learning (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper