ConicFlow: constrained Inference-time optimal control for flow matching models

Key points

• We discretize this problem and adapt receding-horizon control, optimizing only a short window of the trajectory at each step for computational tractability.1
• We further propose a terminal-estimate variant that avoids accumulating local truncation errors by evaluating the reward and constraints at the predicted terminal point rather than at the end of the current horizon.1
• Under convex-affine structure, we derive a sharper bound that scales linearly with omitted horizon length, and we show that the terminal-estimate variant is provably tighter when the flow dynamics change slowly.1
• Experiments in both low- and high-dimensional settings with a recent large-scale flow matching model show that ConicFlow achieves high reward and strong constraint satisfaction while deviating less from the pretrained flow than recent baselines.1

Sources (1)

• [1]ConicFlow: constrained Inference-time optimal control for flow matching models
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 08:32 PM

  We propose ConicFlow, which frames constrained sampling from a flow matching model at inference time as an optimal control problem that balances deviation from the pretrained trajectory against reward, subject to hard terminal constraints.

  We discretize this problem and adapt receding-horizon control, optimizing only a short window of the trajectory at each step for computational tractability.

Extractive summary: sentences quoted from the sources.