POP: Prior-fitted Optimizer Policies

2026-02-17

AI Summary

This work addresses the sensitivity of traditional gradient-based optimizers to hyperparameters and their limited generalization in highly non-convex problems. The authors propose POP, a meta-learned optimizer that achieves strong out-of-the-box generalization without task-specific hyperparameter tuning by leveraging large-scale synthetic priors encompassing both convex and non-convex objectives. POP dynamically predicts per-coordinate step sizes through a context-aware policy network that exploits information from the optimization trajectory. Evaluated across 47 benchmark functions of varying complexity, POP consistently outperforms first-order gradient methods, evolutionary strategies, Bayesian optimization, and existing meta-learning optimizers.

Technology Category

• Search and Optimization: Non-convex Optimization
• Machine Learning: Optimization
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience

Abstract

Optimization refers to the task of finding extrema of an objective function. Classical gradient-based optimizers are highly sensitive to hyperparameter choices. In highly non-convex settings their performance relies on carefully tuned learning rates, momentum, and gradient accumulation. To address these limitations, we introduce POP (Prior-fitted Optimizer Policies), a meta-learned optimizer that predicts coordinate-wise step sizes conditioned on the contextual information provided in the optimization trajectory. Our model is learned on millions of synthetic optimization problems sampled from a novel prior spanning both convex and non-convex objectives. We evaluate POP on an established benchmark including 47 optimization functions of various complexity, where it consistently outperforms first-order gradient-based methods, non-convex optimization approaches (e.g., evolutionary strategies), Bayesian optimization, and a recent meta-learned competitor under matched budget constraints. Our evaluation demonstrates strong generalization capabilities without task-specific tuning.

Problem

• optimization
• hyperparameter sensitivity
• non-convex optimization
• generalization
• gradient-based optimizers

Innovation

• meta-learning
• optimizer policies
• coordinate-wise step sizes
• optimization trajectory
• non-convex optimization