The paper introduces the Energy-based Feynman-Kac Corrector (EBFKC), a method that improves inference-time scaling for diffusion models that are imperfect due to data and training limitations. EBFKC derives Feynman-Kac dynamics to exactly track a prescribed path in continuous time, then approximates these dynamics with sequential Monte Carlo and variance‑controlling guidance. Experiments on Gaussian mixtures, particle systems, and molecular simulations demonstrate that EBFKC closely matches target distributions and free‑energy profiles, outperforming standard inference‑time scaling baselines that still exhibit significant sampling errors.
By Zuokai Wen, Louis Grenioux, Weinan E, Jiequn Han
arXiv:2607. 04780v1 Announce Type: cross Abstract: Sequential Monte Carlo (SMC) methods are a natural tool for post-hoc conditioning of pretrained generative models, but in many applications the mutation kernels used by the particle system are biased approximations of an ideal Feynman--Kac flow.
By Stanislas Strasman (SU, LPSM), Gabriel Victorino Cardoso (LPSM), Sylvain Le Corff (LPSM), Vincent Lemaire (LPSM), Antonio Ocello
Sequential Monte Carlo (SMC) methods are a natural tool for post-hoc conditioning of pretrained generative models, but in many applications the mutation kernels used by the particle system are biased approximations of an ideal Feynman--Kac flow. This paper develops a non-asymptotic error analysis for such SMC samplers.
arXiv:2503. 14549v3 Announce Type: replace-cross Abstract: How can a cheap but biased sequential, finite-horizon sampler over a discrete space be corrected so that its terminal output follows a prescribed Gibbs distribution?
By Michael Chertkov, Sungsoo Ahn, Hamidreza Behjoo
arXiv:2609.37227v1 Announce Type: new
Abstract: Inference-time steering adapts pretrained diffusion and flow-based models to new tasks, e.g., to generate samples from a conditional distribution or sa...
By Adhithyan Kalaivanan, Zheng Zhao, Jens Sj\"olund, Fredrik Lindsten
arXiv:2507. 08390v5 Announce Type: replace Abstract: Discrete diffusion models have recently emerged as strong alternatives to autoregressive language models, matching their performance through large-scale training.
By Meihua Dang, Jiaqi Han, Minkai Xu, Kai Xu, Akash Srivastava, Stefano Ermon
arXiv:2608. 11544v1 Announce Type: cross Abstract: We propose CVaR-penalized Generative Particle Algorithm (CVaR-GPA), a robust, tail-agnostic algorithm for fine-tuning generative models to learn heavy-tailed distributions and capture extreme events, requiring no prior knowledge or estimation of the target's tail characteristics.
By Thejani Gamage, Hyemin Gu, Zhizhen Zhang, Ziyu Chen, Markos Katsoulakis, Luc Rey-Bellet
The paper introduces MM‑SOLD, a training‑free sampling method that uses moment‑matched score‑smoothed overdamped Langevin dynamics to generate samples without training a neural diffusion model. It enforces target moments during sampling, proving convergence to a Gibbs–Boltzmann density that matches the mean and covariance of a training Gaussian mixture model, even for rank‑deficient data. Experiments demonstrate fast, robust latent‑space image generation on CPUs with fidelity and diversity comparable to neural diffusion baselines.
By Zhenyu Yao, Daniel Paulin
arXiv:2503. 14549v4 Announce Type: replace Abstract: Scientific generative models must turn tractable local decisions into globally correlated samples that respect physical constraints.
By Michael Chertkov, Hamidreza Behjoo, Sungsoo Ahn
arXiv:2608. 20123v1 Announce Type: cross Abstract: We study inference-time control for text generation in discrete diffusion language models, where the goal is to steer sampling toward sequence-level rewards without retraining.
By Lohithsai Yadala Chanchu, Hany Abdulsamad, Christian A. Naesseth
We study the problem of sampling from target distributions whose potentials are simultaneously non-smooth, subject to superlinear gradient growth, and non-convex. We introduce the Subgradient Tamed Unadjusted Langevin Algorithm (SG-TULA), a discretisation of the Langevin diffusion that operates directly on subgradients, without relying on computationally demanding smoothing procedures.
ENCORE is a new exact parallel control method for inference‑time steering of pretrained generative models toward tilted target distributions. It uses replica exchange with each replica storing its trajectory, avoiding simulation of the intractable time reversal and ensuring target invariance. Experiments on synthetic, biomolecular, and image data show that ENCORE achieves competitive accuracy and diversity, remains robust to sampler perturbations, and works with distilled samplers where prior replica‑exchange corrections fail.
By Jiahao Yu, Saifuddin Syed, Jos\'e Miguel Hern\'andez-Lobato, Jiajun He