DiFA: Inference-Time Forward-Process Alignment for Diffusion Models

Hugging Face Daily Papers Papers

Summary

Proposes DiFA, a training-free framework that reframes inference-time data prediction refinement as sequential state estimation using Kalman filtering, significantly improving generative fidelity on CIFAR-10 and ImageNet.

The prevailing inference framework for diffusion models formulates generation fundamentally as a problem of numerical integration. This perspective casts the model as an exact estimator, neglecting the inherent statistical uncertainty of the denoising process. In this work, we propose Forward-Process Aligned Diffusion prediction (DiFA), a training-free framework that reframes inference-time data prediction refinement as a sequential state estimation problem. Rather than reusing past outputs solely for numerical integration, DiFA treats iterative data predictions along the reverse trajectory as correlated observations to build a forward-aligned temporal consensus. Inspired by Kalman filtering, this consensus aggregates historical predictions according to structural consistency and noise-level compatibility. To counteract the over-smoothing tendency of temporal consensus, we introduce a deviation guidance mechanism to adaptively preserve residual details. Empirically, DiFA yields significant improvements on CIFAR-10 and ImageNet across the evaluated metrics, including FID, IS, and FD-DINOv2, demonstrating that aligning inference with the forward statistical structure substantially improves generative fidelity.
Original Article
View Cached Full Text

Cached at: 07/21/26, 06:35 AM

Paper page - DiFA: Inference-Time Forward-Process Alignment for Diffusion Models

Source: https://huggingface.co/papers/2607.17972

Abstract

Theprevailinginferenceframeworkfordiffusionmodelsformulatesgenerationfundamentallyasaproblemofnumericalintegration.Thisperspectivecaststhemodelasanexactestimator,neglectingtheinherentstatisticaluncertaintyofthedenoisingprocess.Inthiswork,weproposeForward-ProcessAlignedDiffusionprediction(DiFA),atraining-freeframeworkthatreframesinference-timedatapredictionrefinementasasequentialstateestimationproblem.Ratherthanreusingpastoutputssolelyfornumericalintegration,DiFAtreatsiterativedatapredictionsalongthereversetrajectoryascorrelatedobservationstobuildaforward-alignedtemporalconsensus.InspiredbyKalmanfiltering,thisconsensusaggregateshistoricalpredictionsaccordingtostructuralconsistencyandnoise-levelcompatibility.Tocounteracttheover-smoothingtendencyoftemporalconsensus,weintroduceadeviationguidancemechanismtoadaptivelypreserveresidualdetails.Empirically,DiFAyieldssignificantimprovementsonCIFAR-10andImageNetacrosstheevaluatedmetrics,includingFID,IS,andFD-DINOv2,demonstratingthataligninginferencewiththeforwardstatisticalstructuresubstantiallyimprovesgenerativefidelity.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.17972

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.17972 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.17972 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.17972 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Temporal Difference Learning for Diffusion Models

arXiv cs.LG

This paper introduces a temporal difference (TD) learning objective for diffusion models that enforces cross-time consistency along the denoising trajectory. It reformulates denoising as a reinforcement learning policy evaluation problem, showing significant improvements in sample quality (FID), especially for few-step samplers.