Search papers, labs, and topics across Lattice.
This paper introduces Three-Body Scattering Modeling (TBSM) as a novel approach for generative modeling that leverages a distributional energy to facilitate sample-level motion and regression supervision for a one-step generator. By employing a mechanism where each projectile is attracted to a real source and repelled from a generated source, TBSM achieves significant improvements in generative performance, evidenced by FID scores of 2.23 for pixel-space PixelDiT-XL and 1.63 for latent-space DiT-XL on ImageNet-256. The methodology reduces field noise and enhances the efficiency of training by utilizing a constant-size interaction per projectile, contrasting with traditional methods that rely on all-pairs interactions across batches.
TBSM achieves state-of-the-art FID scores by transforming generative modeling through a novel energy-based interaction mechanism that enhances sample-level generation efficiency.
Modern generative models typically rely on an adversarial critic, a prescribed noise-to-data path, or an autoregressive factorization. Instead, we show that a proper distributional energy can induce sample-level motion and provide direct regression supervision for a one-step generator. Three-Body Scattering Modeling (TBSM) for generation turns the energy distance into a constant-size per-projectile interaction: each projectile is attracted toward one real source and repelled from one independently generated source. Conditioned on the projectile and its condition, its expectation equals the $2$-Wasserstein gradient-flow velocity of $\frac12D_E^2(P_{\theta},Q)$. A batch of $B$ frozen-target events yields $O(B)$ sample-level losses, each using one reference for its condition instead of the minibatch-wide all-pairs field used by methods such as Drifting Models. Tracking this conditional expectation online can reduce field noise. Using scattering in frozen image features, TBSM trains one-step generators on ImageNet-256, achieving FID${}=2.23$ with pixel-space PixelDiT-XL and FID${}=1.63$ with latent-space DiT-XL at NFE${}=1$. We provide a design map relating diffusion-related supervision, Drift-like dynamics, and GAN-like objectives. These results establish tracked scattering as a route to high-dimensional one-step generation. Code: https://github.com/sp12138/TBSM.
