Likelihood Score Approximation (LSA), a generative framework that keeps a pretrained unconditional model fixed and learns an observation-conditioned model that approximates the likelihood score from paired samples, is introduced.
Abstract
Generative models have found great success as data-driven methods of solving inverse problems. Two popular approaches work either by combining a pretrained generative prior with a known degradation model, or by training a conditional generative model directly from paired data. We target a setting that spans both regimes: unknown degradations can be learned from few paired examples, while known degradations can be learned from self-generated samples. We introduce Likelihood Score Approximation (LSA), a generative framework that keeps a pretrained unconditional model fixed and learns an observation-conditioned model that approximates the likelihood score from paired samples. Within a conditional stochastic-interpolant framework, LSA can be trained in either score or velocity coordinates, independently of the unconditional model's native parameterization, and supports both deterministic and stochastic sampling. We further show empirically that the prior model can be swapped post-training while keeping the same LSA model. Across speech and image inverse problems, LSA operates effectively even at roughly 0.01% of the full training dataset. On the ImageNet-256 benchmark it achieves competitive or better restoration quality than strong posterior-sampling baselines while requiring up to several orders of magnitude fewer network evaluations.
Joint twist-flow is proposed, an augmented flow-matching formulation that learns a continuous transport from the augmented source state to the augmented terminal state, helping reduce likelihood-inconsistent variation while preserving variability in weakly constrained directions.
Shi-Qin Zeng, Zi-Jun Deng, Felix J. Herrmann· 0 citations
Variational inference has emerged as a computationally efficient framework for solving Bayesian inverse problems. Rather than relying on random sampling in high-dimensional model spaces, it reformulates probabilistic inference as an optimization problem, marking a fundamental shift from sampling-based to optimization...
Jia-Jia Sun, Si-Hong Wu· The Leading Edge· 0 citations
PRODiGI, a pretrained data-to-program model that infers an explicit, executable generative program in a single forward pass, is introduced and program-space fine-tuning is introduced, which refines differentiable program parameters by matching generated and empirical samples while keeping model parameters intact.
Simon Klüttermann, Xue-Ying Ding, L. Akoglu· 0 citations
This work proposes a method to sample from models'perceptual prior distributions directly, by steering a generative model to produce stimuli along controllable axes and running Gibbs sampling over that space with the model under study as the judge, and recovers both canonical biases and surprising novel priors invisibl...
Manuel Cherep, Pattie Maes, Nikhil Singh· 2 citations
This work develops a framework for source-space generalized Bayesian inference that combines inexpensive few-step prior transports with posterior stability guarantees and establishes Wasserstein error bounds between the exact and learned posteriors in terms of the joint population iMF and auxiliary-velocity loss.
Hoang Phuc Hau Luu, Marcelo Hartmann, Zhong-Jian Wang· 0 citations
This work describes the conditions under which latent-space inference recovers the desired target posterior and systematically study its empirical trade-offs, comparing latent and standard estimators while controlling for network capacity, regularization, optimization, and training compute.
L. Kühmichel, Stefan T. Radev, B. Koppolu et al.· arXiv.org· 0 citations
Related blog posts
MIT News · Artificial Intelligence· news.mit.eduNov 17, 2026
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026