Xiaomi and Universities Propose A-RSI for Recursive Self-Improvement

Xiaomi Research and university collaborators proposed A-RSI, which adaptively refreshes training tasks to fix stale task pools in recursive self-improvement, requiring no labeled data and raising the average score across 11 benchmarks from 57.77 to 61.05.

2026-10-30 ~ 2026-11-01 · 2 related posts

• Xiaomi's A-RSI Boosts Zero-Data Multimodal Self-Improvement via Adaptive Task Refresh — xiaomi-research · 2026-10-30
• Xiaomi's A-RSI lifts 11-benchmark average from 57.77 to 61.05 via adaptive task refresh — mark_k · 2026-11-01