Looped Flows Paper Hits 58.8% on ARC-AGI-1 by Training Recurrence With Denoising Objectives
chaumian · x · 2026-09-13
A new paper proposes looped flows, letting models spend more compute at inference by recurrently updating a hidden state.
- Problem: Looped models typically backprop through only one or a few updates, so early updates fail to support later ones
- Method: Train the recurrence with local denoising objectives, imposing temporal association via progressively decreasing noise levels and shared noise, incentivizing recurrent states that transfer useful computation over time
- Inference: Formulated as integrating the velocity of a probability flow parameterized by the learned denoiser, coupled with recurrent states; finer temporal grids mean more compute, and different initial noise samples yield multiple valid predictions
- Results: Outperforms prior SOTA looped models overall across six reasoning benchmarks, with 58.8% test accuracy on ARC-AGI-1 and 12.2% on ARC-AGI-2
Related event: Looped Flows: Denoising-Trained Recurrent Reasoning Tops Loop-Model ARC-AGI(6 posts)→
More from Research
- Schmidhuber revisits history: 1971 deep network modeled the British economy — SchmidhuberAI · 2026-09-20
- Sort samples by loss and check the head: a practical labeling-error hack — antoine_chaffin · 2026-09-20
- Qwen researcher: human annotation may be noisier than auto-labeling — antoine_chaffin · 2026-09-20
- Retrieval eval datasets are error-prone; LLM relabeling proposed as fix — CShorten30 · 2026-09-20
- PhyFilter from Beihang and NTU lets sim-trained quadrupeds walk real terrain, in npj Robotics — jiqizhixin · 2026-09-20
- ICLR submissions triple to 60,000 abstracts; reviewing would take 250 human-years — gleech · 2026-09-20