Test-Time Meta-Adaptation with Self-Synthesis
This addresses the challenge of test-time adaptation for LLMs across diverse domains, offering a novel approach to self-improvement, though it appears incremental as it builds on existing meta-learning and synthetic data methods.
The paper tackles the problem of enabling large language models to adapt and self-improve at test time by introducing MASS, a meta-learning framework that generates problem-specific synthetic training data for targeted self-updates, resulting in effective and data-efficient adaptation in mathematical reasoning tasks.
As strong general reasoners, large language models (LLMs) encounter diverse domains and tasks, where the ability to adapt and self-improve at test time is valuable. We introduce MASS, a meta-learning framework that enables LLMs to self-adapt by generating problem-specific synthetic training data and performing targeted self-updates optimized for downstream performance at inference time. We train this behavior end-to-end via bilevel optimization: an inner loop adapts on self-generated examples while an outer loop meta-learns data-attribution signals and rewards post-update task performance. The synthetic data is optimized with scalable meta-gradients, backpropagating the downstream loss through the inner updates to reward useful generations. Experiments on mathematical reasoning show that MASS learns to synthesize per-instance curricula that yield effective, data-efficient test-time adaptation.