A Kinetic Energy Perspective of Flow Matching
This work provides a new diagnostic and inference strategy for researchers and practitioners working with flow-based generative models, addressing issues of generation quality and memorization.
This paper introduces Kinetic Path Energy (KPE), a per-sample diagnostic for flow-based generative models that measures accumulated kinetic effort along an ODE trajectory. They found that higher KPE predicts stronger semantic fidelity and that high-KPE trajectories land in sparse representation regions, but paradoxically, extreme energies lead to memorization. They propose Kinetic Trajectory Shaping (KTS), a training-free inference strategy that improves generation quality and reduces memorization.
Flow-based generative models can be viewed through a physics lens: sampling transports a particle from noise to data by integrating a learned velocity field, and each sample corresponds to a trajectory with its own dynamical effort. Motivated by classical mechanics, we introduce Kinetic Path Energy (KPE), an action-like, per-sample diagnostic that measures the accumulated kinetic effort along an ordinary differential equation (ODE) trajectory. Empirically, KPE exhibits two robust correspondences: {i} higher KPE predicts stronger semantic fidelity; {ii} high-KPE trajectories land in sparse representation regions. We further provide theoretical guarantees linking trajectory energy to data sparsity. Paradoxically, this correlation is non-monotonic. At sufficiently high energy, generation can degenerate into memorization. Leveraging the closed-form formula of empirical flow matching, we show that extreme energies drive trajectories toward near-copies of training examples. This yields a Goldilocks principle and motivates Kinetic Trajectory Shaping (KTS), a training-free two-phase inference strategy that boosts early motion and enforces a late-time soft landing, reducing memorization and improving generation quality across benchmark tasks.