Unrolled Flow Models for Reasoning

Key points

• We prove that a flow parameterized by a two-layer Transformer can solve graph reachability, with the required number of integration steps increasing with the target's distance from the root.1
• Yet, standard flow language models can fail to benefit from additional steps on reasoning tasks.1
• To address this, we instead train through the model's own latent rollout over a randomly sampled subinterval of [0, 1], decoding only at the endpoint.1
• Together, these results establish a theoretical basis for reasoning with flows and show how rollout training, stable latent dynamics, and rollout selection help realize this capacity in practice.1

Sources (1)

• [1]Unrolled Flow Models for Reasoning
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 09:45 AM

  Flow matching enables language generation in few steps, but whether additional integration steps improve reasoning remains unclear.

  We prove that a flow parameterized by a two-layer Transformer can solve graph reachability, with the required number of integration steps increasing with the target's distance from the root.

Extractive summary: sentences quoted from the sources.