Memory Transfer Learning: How Memories are Transferred Across Domains in Coding Agents

arXiv cs.AI / 4/16/2026

📰 NewsSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper proposes Memory Transfer Learning (MTL) to let coding agents use a unified memory pool across heterogeneous coding domains rather than limiting memory to a single task type.
Experiments on six coding benchmarks show cross-domain memory improves average performance by 3.7%, with gains driven mainly by transferable meta-knowledge (e.g., validation routines) instead of task-specific code.
The authors find that the abstraction level of memory determines transferability: high-level insights generalize well, while low-level concrete traces can cause negative transfer due to over-specificity.
Transfer effectiveness increases as the memory pool grows, and the approach can transfer memory even between different agent/model architectures.
The work provides empirical design principles for expanding memory utilization beyond single-domain “memory silos,” and points to a project page for further details.

Abstract

Memory-based self-evolution has emerged as a promising paradigm for coding agents. However, existing approaches typically restrict memory utilization to homogeneous task domains, failing to leverage the shared infrastructural foundations, such as runtime environments and programming languages, that exist across diverse real-world coding problems. To address this limitation, we investigate \textbf{Memory Transfer Learning} (MTL) by harnessing a unified memory pool from heterogeneous domains. We evaluate performance across 6 coding benchmarks using four memory representations, ranging from concrete traces to abstract insights. Our experiments demonstrate that cross-domain memory improves average performance by 3.7\%, primarily by transferring meta-knowledge, such as validation routines, rather than task-specific code. Importantly, we find that abstraction dictates transferability; high-level insights generalize well, whereas low-level traces often induce negative transfer due to excessive specificity. Furthermore, we show that transfer effectiveness scales with the size of the memory pool, and memory can be transferred even between different models. Our work establishes empirical design principles for expanding memory utilization beyond single-domain silos. Project page: https://memorytransfer.github.io/