Learning to Order: Task Sequencing as In-Context Optimization

2026-03-15

AI Summary

This work addresses the limited few-shot generalization capability of task sequencing in novel scenarios—such as robotic assembly and autonomous driving—by proposing a meta-learning-based solution. The approach leverages large-scale synthetically generated task sequences modeled as paths in directed graphs, enabling effective meta-learning under an unbounded task prior for the first time. Employing a Transformer-based architecture, the model learns to rapidly infer optimal execution orders from only a few demonstrations by treating task sequences as graph traversal problems. Experimental results demonstrate that the proposed method significantly outperforms non-meta-learning baselines, achieving superior efficiency in discovering optimal task sequences under few-shot conditions and thereby validating its strong generalization ability.

Technology Category

• Search and Optimization: Metareasoning and Metaheuristics
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling
• Intelligent Robots: Learning & Optimization for ROB

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Task sequencing (TS) is one of the core open problems in Deep Learning, arising in a plethora of real-world domains, from robotic assembly lines to autonomous driving. Unfortunately, prior work has not convincingly demonstrated the generalization ability of meta-learned TS methods to solve new TS problems, given few initial demonstrations. In this paper, we demonstrate that deep neural networks can meta-learn over an infinite prior of synthetically generated TS problems and achieve a few-shot generalization. We meta-learn a transformer-based architecture over datasets of sequencing trajectories generated from a prior distribution that samples sequencing problems as paths in directed graphs. In a large-scale experiment, we provide ample empirical evidence that our meta-learned models discover optimal task sequences significantly quicker than non-meta-learned baselines.

Problem

• Task Sequencing
• Few-shot Generalization
• Meta-learning
• Deep Learning

Innovation

• task sequencing
• meta-learning
• in-context optimization
• transformer architecture
• few-shot generalization