Model adapts task refresh to boost learning without external data
A-RSI: Adaptive Task Refresh for Zero-Data Multimodal Self-Improvement
Artificial Intelligence
Summary
Machine learning models improve by learning from new examples, but it's hard to know when to introduce fresh learning tasks. The authors present a system called A-RSI that monitors the model's training signals and decides when to refresh tasks automatically. It does this without relying on human-labeled answers or external data, using its own generated tasks and verifying results independently. This approach led to better performance across multiple test benchmarks.
What this means in practice
- •For machine learning engineers: Improve model training efficiency by automatically refreshing task sets based on reward signals without requiring external labeled data.
- •For automated testing teams: Generate diverse multimodal test questions and answers programmatically to audit AI system capabilities without manual labeling.
Authors
Yu Liu, Youxing Li, Yuedong Tan, Zhiwei Yang, Zihan Su, Hanqi Feng, Shaojie Zhang, Libin Liao, Zihao Wan, Chenxu Wu, Cong Cao, Fangfang Yuan, Yanbing Liu, Ziheng Chen, Pei Fu, Radu Timofte, Zongwei Wu, Nicu Sebe, Barnabas Poczos, Jian Luan
Abstract
Recursive self-improvement requires models not only to acquire useful training experience, but also to decide when their current experience has ceased to be useful. Existing methods often focus on what experience to generate next, while paying less attention to when fresh experience should be introduced. Many also construct training targets through self-consistency or majority voting, making supervision depend on the model's own uncertain predictions. We address these two questions separately: training dynamics determine when to acquire new experience, while the answers to new tasks remain independently checkable. A-RSI monitors reward variation and accuracy in training logs to refresh a task pool when its learning signal declines, without consulting validation or test scores. A static library of model-authored programs produces multimodal questions and derives their reference answers; we specify the task-generation contract, automatically verify chosen properties of the reward and refresh rules, and separately audit the generated answers. The training loop uses no external training examples or human-written answer labels. On the primary eleven-benchmark suite, A-RSI raises the performance average from 57.77 to 61.05 with 800 fresh tasks.