Learning to Learn-at-Test-Time: Language Agents with Learnable Adaptation Policies
This work addresses the need for more effective and generalizable adaptation strategies in language agents, offering a novel approach to automate policy optimization rather than relying on human intuition, though it is incremental in advancing TTL methods.
The paper tackles the problem of hand-crafted adaptation policies in Test-Time Learning (TTL) for language agents by introducing Meta-TTL, a framework that learns optimal adaptation policies through bi-level optimization, resulting in consistent performance improvements over baselines on Jericho and WebArena-Lite benchmarks in both in-distribution and out-of-distribution settings.
Test-Time Learning (TTL) enables language agents to iteratively refine their performance through repeated interactions with the environment at inference time. At the core of TTL is an adaptation policy that updates the actor policy based on experience from previous episodes, thereby improving future behavior. Existing methods rely on fixed, hand-crafted adaptation policies rather than optimizing them for downstream improvement. We argue that optimal adaptation policies should be learned from task environments, not hand-engineered based on human intuition. To achieve this, we introduce Meta-TTL, a framework that formulates the discovery of effective adaptation policies as a bi-level optimization problem. Within this framework, the inner loop executes the standard TTL process, measuring how effectively a candidate adaptation policy helps an agent correct errors across sequential episodes. Guided by the agent's performance, the outer loop employs evolutionary search over a diverse distribution of training tasks to iteratively refine the adaptation policy. We evaluate Meta-TTL on Jericho and WebArena-Lite across both in-distribution (ID) and out-of-distribution (OOD) settings, using multiple meta-agent backbones. Results on both benchmarks show that Meta-TTL consistently outperforms hand-crafted baselines, suggesting that the optimized adaptation policy encodes transferable strategies that generalize beyond the training task distribution.