Reflective Context Learning: Studying the Optimization Primitives of Context Space

2026-04-03

AI Summary

Existing in-context learning approaches lack systematic solutions to challenges such as credit assignment, overfitting, catastrophic forgetting, local optima, and high variance. This work proposes Reflective Contextual Learning (RCL), a novel framework that, for the first time, systematically introduces classical optimization primitives into contextual space. RCL enables agents to optimize their learning through interaction with the environment, reflection on actions and failures, and iterative refinement of context representations. The framework unifies prior methods while incorporating mechanisms including directional update signals, contextual perturbation, batch processing, failure replay, trajectory grouping, curriculum strategies, and variants of optimizer states. Experimental results demonstrate that RCL significantly outperforms strong baselines on AppWorld, BrowseComp+, and RewardBench2, validating both the efficacy and cross-task transferability of its constituent optimization primitives.

Abstract

Generally capable agents must learn from experience in ways that generalize across tasks and environments. The fundamental problems of learning, including credit assignment, overfitting, forgetting, local optima, and high-variance learning signals, persist whether the learned object lies in parameter space or context space. While these challenges are well understood in classical machine learning optimization, they remain underexplored in context space, leading current methods to be fragmented and ad hoc. We present Reflective Context Learning (RCL), a unified framework for agents that learn through repeated interaction, reflection on behavior and failure modes, and iterative updates to context. In RCL, reflection converts trajectories and current context into a directional update signal analogous to gradients, while mutation applies that signal to improve future behavior in context space. We recast recent context-optimization approaches as instances of this shared learning problem and systematically extend them with classical optimization primitives, including batching, improved credit-assignment signal, auxiliary losses, failure replay, and grouped rollouts for variance reduction. On AppWorld, BrowseComp+, and RewardBench2, these primitives improve over strong baselines, with their relative importance shifting across task regimes. We further analyze robustness to initialization, the effects of batch size, sampling and curriculum strategy, optimizer-state variants, and the impact of allocating stronger or weaker models to different optimization components. Our results suggest that learning through context updates should be treated not as a set of isolated algorithms, but as an optimization problem whose mechanisms can be studied systematically and improved through transferable principles.

Problem

• context space
• optimization primitives
• credit assignment
• overfitting
• learning signals

Innovation

• Reflective Context Learning
• context space optimization
• optimization primitives
• credit assignment
• variance reduction