Denoising
-
Shared-Factor-Conditioned Diffusion for Low-Rank Tensor Recovery
2026-10-26 12:53:53Zhengyun Cheng, Guanwen Zhang, Wei ZhouAbstract
Tensor recovery spans heterogeneous data structures and observation processes, calling for a representation that can adapt across modalities and tasks. We propose DiffTR, an unsupervised framework that unifies tensor completion, denoising, and hyperspectral super-resolution through coordinate-conditioned diffusion. Shared factors indexed by tensor coordinates condition a compact score network, and score matching jointly learns the factors and an implicit tensor representation from the available observations. Scalar and vector-valued outputs accommodate individual entries and entire tubes, enabling a common formulation for tabular tensors, images, and videos. The framework separates representation learning from task-specific observation modeling: reverse diffusion recovers missing values; estimation via Tweedie's formula, soft thresholding, and optional spatial consistency support mixed-noise denoising; and measurement-guided sampling transfers a prior learned from low-resolution hyperspectral observations to high-resolution reconstruction, followed by post-sampling refinement with data consistency and spectral subspace regularization. This separation makes the representation compatible with different corruption models and permits task-specific constraints to be incorporated within the same recovery framework. Experiments across multiple tensor modalities and recovery tasks demonstrate competitive reconstruction quality and the flexibility of a unified conditional generative representation. The code is available at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.35368
PDF
-
Low-rank MERA Decomposition of 3D OCT Image Volume for Compression, Denoising, Feature Extraction and Diagnosis of Retinal Diseases
2026-10-26 11:45:42Ivica Kopriva, Fei Shi, Marija \v{S}tanfel, Qin Peng, Jianhao Bao, Gaowei Li, Xinjian ChenAbstract
Optical coherence tomography (OCT) is a non-invasive imaging modality used for visualization of retinal tissue. However, OCT volumes are affected by speckle noise, which degrades image contrast and obscures fine structural details. Furthermore, modern clinical practice requires large field-of-view, high-resolution OCT volumes, whose storage and transmission are demanding for archiving and teleophthalmology applications. To address aforementioned challenges, we propose a low-rank multi-scale entanglement renormalization ansatz (MERA) tensor network approximation of 3D OCT image volume, where the image volume is represented as contractions of a top core factor encoding global features and multiple interacting low-rank orthogonal or semi-orthogonal factors modeling inter- and intra-B scan correlations. This representation jointly enables compression, denoising, and feature extraction. Unlike deep learning approaches, the proposed approach is training-free and processes each OCT volume independently. We evaluate the proposed framework on OCT image volumes acquired by two different scanners and demonstrate improved visual quality and reduced retinal layer segmentation errors in comparison with several competing methods. We also extract MERA features from two clinical OCT volume datasets. They are subsequently classified using a Grassmann-manifold nearest-subspace classifier. With the subspace dimension as its only hyperparameter, the classifier is theoretically shown to be robust in small-sample clinical scenario. We demonstrate competitive accuracy of classification of retinal diseases in these small-sample settings, where deep learning methods are often limited by the scarcity of labeled data. The proposed structure-aware model-driven representation learning for high-dimensional data under limited supervision provides good generalization performance while maintaining high interpretability.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35271
PDF
-
DTCA: Denoising Token Credit Assignment for Reinforcement Learning in Diffusion Language Models
2026-10-26 10:54:11Hongze Chen, Miao Pan, Lizhou Liang, Qinfeng Li, Xinyin Ma, Yongliang Shen, Weishi Wang, Daniel Dahlmeier, Xuhong Zhang, Wenqi ZhangAbstract
Diffusion language models (DLMs) generate text through iterative denoising, enabling bidirectional context modeling and parallel token refinement. Recent studies have applied reinforcement learning (RL) to DLMs and improved their reasoning performance. However, the iterative denoising process makes credit assignment more challenging, as final outcome rewards provide limited information about how different tokens contribute to the final output. In this paper, we propose Denoising Token Credit Assignment (DTCA), a framework for token-level credit assignment in reinforcement learning for diffusion language models. DTCA estimates token contributions by measuring the improvement in prediction confidence of remaining masked positions after token commitment and groups these signals according to part-of-speech (POS) categories. These contribution weights are used to select informative denoising timesteps and assign token-level learning signals during RL training. DTCA further calibrates these learning signals according to token-level entropy changes induced by policy updates. Experiments on LLaDA-8B-Instruct show that DTCA consistently improves mathematical reasoning performance across different generation budgets, with gains of up to 9.0 points on MATH500 and 5.3 points on GSM8K over the base model.
Abstract (translated)
URL
https://arxiv.org/abs/2610.35201
PDF
-
PosePlaner: Denoising Any Feed-Forward Pose Predictor from Pairwise Planar Geometry
2026-10-26 04:55:36Yihang Chen, Bin Tan, Yujun Shen, Yinghao Xu, Nan XueAbstract
Recent multi-view Transformers (e.g., DUSt3R and VGGT) have advanced 3D reconstruction by scaling training data and model capacity from two views to many views and long sequences, subsuming the classical SfM/SLAM pipeline of feature extraction, matching, and pose estimation in a single feed-forward pass. Yet this subsumption hides the interfaces between pipeline stages---correspondences, geometry, and pose are entangled in a single forward pass, leaving supervised fine-tuning as the major path to improvement and offering no mechanism to correct predictions from geometric evidence at test time. We propose PosePlaner to close this gap by reviving the classical coupling between correspondences and pose via flow matching, treating pose refinement as a conditional generation problem that takes coarse geometric matches and initial pose as input condition and fuses the resulting pose hypotheses into a single refined estimate, providing a continuous, data-driven analog of RANSAC voting. Our method is trained efficiently on two-view data, yet serves as a periodic refiner for streaming n-view pose estimators with no retraining. Empirically, it corrects failures of feed-forward predictors, substantially improves correspondence-based solvers, and reduces accumulated drift in long streaming sequences.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34845
PDF
-
ToDoPro: Faster and Better Reference-Based Generation: Early Token Reduction via Downsampled Self-Probing
2026-10-26 03:27:40Caiyi Sun, Yujing Sun, Siu-Ming YiuAbstract
Recent generic reference-based image generation models natively support multiple conditions by concatenating all reference image and text tokens into a joint sequence processed by a single Diffusion Transformer (DiT). However, this versatile paradigm comes at an efficiency cost: accommodating multiple references leads to a token explosion, incurring massive $\mathcal{O}(N^2)$ computational complexity. Most acceleration methods operate inside the DiT, showing limited capability in reducing spatial redundancy and failing to alleviate peak VRAM. While some pre-model token reduction alternatives exist, they inevitably introduce external models. This raises a question: Is it possible to achieve acceleration and memory savings while preserving compatibility with in-model techniques and requiring no additional parameters? In this work, we realize this goal by proposing the early Token reduction via Downsampled self-Probing (ToDoPro). By executing a few denoising steps on spatially $2\times$ downsampled reference images, this cheap probing stage utilizes the DiT model itself to effectively and efficiently extract reference token importance. Using the probed importance, we apply a hierarchical reduction before the main denoising stage: we completely discard low-importance tokens, substitute mid-importance tokens with their downsampled counterparts to cheaply preserve global context, and keep high-importance tokens. Experiments across three base models and two benchmarks demonstrate that our early reduction accelerates inference significantly and mitigates peak GPU memory. Furthermore, we can improve generation quality by safely filtering out noisy, uninformative reference tokens, enabling the model to concentrate its limited attention capacity on genuinely useful tokens. The code will be released at this https URL
Abstract (translated)
URL
https://arxiv.org/abs/2610.34782
PDF
-
VLA-HWBench: Where the Time Goes When VLA Models Run on Edge Hardware
2026-10-26 02:07:42Borui Xu, Hongshi Tan, Wenqi Jiang, Bingsheng He, Ming Wu, Jason Zeng, Michael HeinrichAbstract
Vision--language--action (VLA) models increasingly run on edge devices aboard robots, where inference latency constrains closed-loop robot control. However, it remains unclear how to design VLA models, their inference software and edge hardware so that robots can run VLA inference in real time. Most VLA models report only task success, while prior studies cover limited action-generation strategies, divide inference into at most three stages, and lack comparisons of inference optimizations on the same this http URL 18 open VLA models across released and alternative backends on NVIDIA Jetson AGX Thor and AMD Ryzen AI Max+ PRO 395, with a seven-stage inference breakdown. Our study reveals three insights. First, action-generation strategy matters more than model size. It dominates 14 of 19 Jetson configurations, latency varies by $82\times$ on Jetson and $157\times$ on AMD, and parameter count explains at most 15% of latency variance. Models that generate long action chunks in a few flow-matching steps keep up with control, whereas token-by-token decoding does not. Second, kernel fusion with 4-bit weights helps most, running $\pi_{0.5}$ and GR00T N1.7 2.6--2.9$\times$ faster than PyTorch with \texttt{this http URL}, but it exists for only these two models and only on the Jetson. Optimization also moves the time to the stages it shortens least, as CPU preprocessing grows from 12\% to 39\% of the GR00T N1.7 call. Third, after optimization, memory bandwidth limits only the stages that process few tokens on the Jetson, such as the denoising steps, whereas software limits all other stages, from unfused kernels to serial preprocessing on the host CPU. Overall, low-latency edge VLAs should use few action-generation iterations, inference software should extend kernel fusion and low-bit weights to more models and platforms, and edge hardware needs high memory bandwidth and a fast host CPU.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34717
PDF
-
XPress: Parallel Refinement for Diffusion Drafters in Speculative Decoding
2026-10-26 00:18:30Zheng Wang, Davis Wertheimer, Yu Chin Fabian Lim, Mudhakar Srivatsa, Raghu K. Ganti, Minjia Zhang, Naigang WangAbstract
Block-diffusion drafters like dFlash generate an entire block of draft tokens in a single forward pass, drastically reducing the overhead of multiple-token drafting in speculative decoding. The crucial final step of the single-pass discrete denoising process involves using the logit distribution at each position to sample conditionally independent tokens. The resulting draft is thus a set of per-position marginals, rather than a joint distribution: no draft token is guaranteed to depend on its predecessors. Such independently sampled marginals tend to produce sequences with tokens that are individually likely, but jointly improbable under the target model's distribution, which verifies each token conditionally. This can cause early rejection and limits acceptance length. To address this, we propose xPress as a means to restore the missing causality in diffusion drafters. xPress is a lightweight causal refiner that reconciles the whole diffusion block at once through parallel refinement, restoring and propagating causal dependencies across the draft without a token-by-token loop. On Qwen3-8B, across seven math, code, and chat benchmarks, xPress raises acceptance length by about 30% on average (up to +56%) and its end-to-end decoding throughput by about 1.3 on average (up to 1.7) compared to the original dFlash diffusion drafter.
Abstract (translated)
URL
https://arxiv.org/abs/2608.02438
PDF
-
EditDistill: Is It Possible to Guide Video Editing with Image Editing?
2026-10-25 14:12:55Guojun Lei, Hong Li, Hongbing Yang, Lixue Gong, Chi Wang, Zheng DongAbstract
Recent advances in image editing have enabled fine-grained and highly controllable visual modifications. However, achieving similar levels of controllability in video editing remains an open challenge, particularly for precise local edits, dynamic temporal effects, and consistent lighting across this http URL this paper, we propose EditDistill, a framework that distills image-level visual transformations from edits to guide video this http URL, given a source video and an edit prompt, we first generate an edited version of the initial video frame using an off-the-shelf image editing model. We then distill the visual transformation of this original-edited image pair into a compact representation termed an edit this http URL descriptor is injected into a video diffusion model through our lightweight fusion module, which steers the denoising process toward the desired edit. To align the video trajectory with the demonstrated image edit while preserving motion and layout, we train with edit-descriptor alignment and DINO structural consistency losses, and derive an equivalent inference-time guidance formulation under flow this http URL experiments demonstrate that our method achieves strong performance on standard benchmarks and produces high-quality video editing results across diverse scenarios. For more details and videos, please refer to the anonymous webpage: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34303
PDF
-
Timing Matters Twice: When to Query and When to Skip in Flow-Matching On-Policy Distillation
2026-10-25 10:00:33Renye Yan, Junyi Zhan, Jikang Cheng, Wei PengAbstract
On-policy distillation (OPD) enables flow-matching models to acquire complementary capabilities from specialized teachers through supervision on student-generated trajectories. However, different evaluation dimensions, such as aesthetics and OCR accuracy, and different tasks, such as image generation and editing, emphasize different aspects of the denoising process, suggesting that their effective supervision may occur at different stages. Existing methods typically select supervision states from predefined timestep distributions, which may overlook these phase-dependent differences and incur unnecessary computational overhead. This motivates two critical timing decisions: when to adaptively query teachers for effective supervision based on the evolving generative process, and when to skip the remaining rollout to avoid redundant computation. In this paper, we introduce PhaseOPD, a phase-aware query-and-skip framework that identifies generative phases online from revisions of predicted-clean latents, using existing student velocity predictions without additional model evaluations. For each task-assigned teacher, PhaseOPD selects a supervision state based on its phase preference and skips the unused rollout suffix once all query decisions are finalized, preserving the single-query velocity-distillation objective. On SD3.5-M, PhaseOPD executes only 5.56 of 16 rollout positions per local batch, reducing training time by 41.3% relative to DanceOPD while improving Aesthetic, GenEval, and OCR scores.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34191
PDF
-
Freeze What Is Settled, Cache What Is Smooth: Spatial Freezing Keeps Step Caching in Its Low-Error Regime
Abstract
Instruction-based image editors built on diffusion transformers run tens of denoising steps. Training-free acceleration either reuses features across steps (step caching) or stops computing the regions an edit leaves untouched (region freezing), and the two have been combined before. We ask how a high speed-up target should be split between them. On Qwen-Image-2.1, freezing does more than remove tokens: it lowers the cost of every step, so the step cache can reach the same overall speed-up with a conservative threshold that keeps its error small. A per-step cost model fitted on development runs predicts held-out latency with a median error of 4.0% and reproduces the speed ceilings of the region methods we re-implemented (2.73$\times$ predicted vs. 2.70$\times$ measured for RegionE). Driving frozen tokens to the model's one-step $x_0$ estimate beats returning them to the reference image on every development sample, and in 90% of measured cases the composed error stays below the sum of the two individual errors. The resulting method, FreezePlan, keeps five full steps and gives the rest of the budget to TaylorSeer. On a pre-registered test split it runs 5.05$\times$ faster end to end at 29.8 dB PSNR to the original output, with no significant drop in a VLM judge score; TaylorSeer at 5.02$\times$ reaches 23.5 dB. The curve extends to 5.99$\times$ at 28.0 dB. Swapping in the region selectors of prior methods under our schedule lowers median PSNR by only 0.5--0.6 dB, which places the gain in the schedule and budget split rather than in the selector.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34177
PDF
-
Learning the Autoregressive: Direct Score-Difference Estimation with Amortized Future Credit for Video Distillation
2026-10-25 06:44:06Zixuan Duan, Yabo Chen, Xunzhi Xiang, Xin Zhang, Haibin Huang, Yin Chen, Chi Zhang, Qi FanAbstract
Few-step distillation for autoregressive (AR) video generation commonly first initializes an AR student and then refines it with distribution matching distillation (DMD). This refinement process faces two challenges. First, although initialization already yields a reasonably capable student, DMD obtains the refinement direction indirectly by estimating the full student score before forming the teacher-student correction. Second, detached-history training prevents the generator from assigning credit to a current chunk for its influence on subsequent chunks. We introduce Leta, a few-step distillation framework that addresses these challenges with ScoreDelta and FutureCredit. ScoreDelta directly estimates the teacher-student correction from denoising residuals, while FutureCredit learns an auxiliary cross-chunk update signal from vector-Jacobian products (VJPs) of next-chunk teacher residuals. Leta combines these signals into a joint teacher-guided update field in a single forward pass during generator updates. Teacher evaluations and credit VJPs are used only for estimator training. For ScoreDelta, we establish population equivalence to score-difference estimation and a conditional approximation advantage under limited capacity in a fixed-feature linear model. In our controlled experiments, Leta improves over DMD by 0.9 points on VBench and 0.66 on VideoAlign. This advantage persists at a higher generator update frequency (a generator-to-estimator update ratio of G:E=1:1), where Leta improves over DMD by 1.6 on VBench and 0.35 on VideoAlign. Project page: this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.34092
PDF
-
TurboWAM: Learning What Matters for Fast World Action Models
2026-10-24 23:04:10Guile Wu, Shuangzhi Li, Xingxin Chen, Chengjie Huang, Sam Hosseini, Tongtong Cao, Yingxue Zhang, Dongfeng BaiAbstract
World action models have demonstrated promising performance in robot policy learning, yet their high inference latency remains a critical bottleneck for real-world deployment. Recent approaches have sought to improve inference efficiency by reducing test-time future-frame imagination or accelerating action prediction. However, these methods often leverage representations from all video world model layers without explicitly considering their varying relevance to action prediction, leading to redundant computation and suboptimal representation utilization. In this work, we propose TurboWAM, a fast world action model that efficiently aggregates structured world representations conditioned on action predictions. Specifically, TurboWAM introduces a structured world representation aggregation module that adaptively fuses representations from different world model layers according to their relevance to the evolving action denoising state. Furthermore, TurboWAM employs a structured action supervision mechanism that adaptively modulates gradients from action supervision across different world model layers during backward propagation, facilitating world representation learning. During inference, by reducing the number of world model layers executed and caching relevant representations across action denoising steps, TurboWAM substantially reduces computational overhead. Extensive experiments on multiple robot manipulation benchmarks demonstrate that TurboWAM achieves low inference latency while maintaining competitive task success rates compared with state-of-the-art methods.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33941
PDF
-
UltraFlash: Accelerating Megapixel Visual Synthesis
2026-10-24 10:19:20Phuc Lai, Anh Nguyen, Phong Nguyen, Anh TranAbstract
Megapixel visual synthesis with latent diffusion models requires operating far beyond the resolutions at which these models are typically trained. Cascade refinement is a leading strategy for this setting: it starts from a base image at a native resolution and progressively upsamples it to the target resolution through multiple refinement stages. Each stage typically performs a pixel-space resolution transition followed by latent-space denoising. While prior acceleration efforts mainly focus on reducing denoising cost, we identify a complementary and underexplored bottleneck: the repeated movement of representations through the VAE. We call these operations pixel-space transitions. In cascade pipelines, intermediate latents are often decoded to RGB, resized, and re-encoded before refinement, while final latents are decoded patch by patch with dense overlap to suppress boundary artifacts. We propose UltraFlash, a transition-efficient approach that accelerates these VAE-heavy operations. UltraFlash has two components. Latent Hyperloop replaces intermediate RGB round trips with a learned latent-space shortcut that approximates the decode-resize-encode transformation without repeated VAE decoding and encoding. Sparse Patch-based Decoding accelerates final reconstruction by addressing patch seams caused by inconsistent GroupNorm statistics and reusing cached statistics across patches, enabling sparse VAE decoding with much smaller overlap. Integrated into a state-of-the-art cascade baseline, UltraFlash reduces 4K generation latency by approximately 2x while improving image quality. These results show that optimizing pixel-space transitions can improve both efficiency and fidelity, offering an effective route toward fast, high-quality megapixel visual synthesis.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33581
PDF
-
How To Build A Modern Diffusion Language Model
2026-10-24 01:50:31Volodymyr Kuleshov, Marianne Arriola, Yair Schiff, Gilad Turok, Guanghan WangAbstract
The frontier of large language model (LLM) performance has advanced rapidly in recent years, fueled by scaling laws for pretraining on large text corpora. However, mainstream LLMs have fundamental limitations: they generate text autoregressively, producing outputs sequentially from left to right one token at a time. This sequential process leaves parallel GPU hardware underutilized during inference--a bottleneck that has grown acute as progress shifts toward scaling post-training and inference-time compute--and commits irreversibly to each token without the ability to revise earlier mistakes. Discrete diffusion models address these limitations by generating tokens in parallel, starting from a noisy or masked sequence and iteratively refining the entire output over a series of denoising steps. This parallel generative process enables speeds of thousands of tokens per second, provides a new axis of inference-time compute scaling, and supports built-in error correction and controllable generation. Over the past two years, discrete diffusion has scaled to billions of parameters across biological sequence modeling and natural language, with open-weights and commercial releases from leading labs. This article provides a survey of key ideas in discrete diffusion and a high-level introduction to how modern diffusion LLMs are built, covering their foundations, core algorithmic and architectural building blocks, applications in biology and language, and implications for future LLM scaling laws.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33391
PDF
-
Matching Particle-Arrival Rates for Low-Count Image Denoising
2026-10-23 15:12:34Huazhen Liu, Zipei Yan, Jizhou LiAbstract
Low-count imaging arises when limited illumination or dose yields only a few detected particles in microscopy, such as photons or electrons, causing measurements to be dominated by signal-dependent Poisson shot noise. Supervised denoisers typically require paired clean and noisy observations, whereas many self-supervised approaches exploit spatial or temporal redundancy without directly estimating the underlying particle-arrival process. We introduce PARM, a self-supervised denoising framework based on particle-arrival-rate matching. By modeling particle accumulation over exposure as a Poisson counting process, PARM identifies the conditional particle-arrival-rate field with the posterior mean of the latent clean rate. Poisson thinning then splits each observation into complementary lower-exposure views that provide self-supervised rate targets for one another. In implementation, the network estimates the normalized spatial allocation of the conditional rate, while the observed total count determines its global particle mass. Experiments on synthetic data, natural-video sequences corrupted with Poisson noise, and real low-dose electron microscopy demonstrate leading or competitive performance while effectively preserving fine structures. PARM therefore provides a statistically grounded approach to low-count image denoising when clean references are unavailable.
Abstract (translated)
URL
https://arxiv.org/abs/2610.33136
PDF
-
SCDM-HTD: Spectral-Wise Conditional Diffusion Model for Hyperspectral Target Detection
2026-10-23 15:08:07Dunbin Shen, Wenfeng Kong, Jianjun Liu, Xiaorui Ma, Hongyu WangAbstract
Hyperspectral target detection (HTD) aims to locate and recognize objects at a fine-grained scale, playing a pivotal role in remote sensing. Despite significant progress, current HTD methods primarily focus on feature comparison while neglecting distribution learning for background-target separation, which can limit detection robustness in practical applications. This paper addresses this issue by introducing a spectral-wise conditional diffusion model that reformulates HTD as a task of generating a detection map. Specifically, we treat background and target confidence scores as training samples and use the corresponding pixel spectra as guiding conditions to capture the distributional discrepancy between background and target through diffusion and denoising. To enhance practical applicability, we propose a lightweight conditional denoising network based on multilayer perceptron to map noisy confidence scores to their ideal counterparts and adopt a deterministic reverse non-Markov chain to skip sampling steps, thereby accelerating the generation process. Furthermore, we introduce a composite loss to improve background-target separation. This loss integrates constrained energy minimization loss to enhance target saliency and suppress the background by maximizing the target response while minimizing average output energy, alongside binary cross-entropy loss to classify background and target using pixel-wise labels. Experimental results on four benchmark datasets demonstrate the effectiveness of the proposed method, with improvements in both quantitative metrics and qualitative visual effects.
Abstract (translated)
URL
https://arxiv.org/abs/2610.32768
PDF
-
LLaCO: Continuous Diffusion for Large Language Models with Competition-Aware Semantic Scheduling
2026-10-23 13:28:40Zijian Jia, Peixuan Zhang, Genying Li, Di Wu, Yujie Fu, Si Li, Boxin Shi, Shuchen WengAbstract
Continuous diffusion models have emerged as a promising non-autoregressive paradigm for text generation over continuous embedding spaces. Along the continuous trajectory, the residual token identity directly shapes the denoising objective. However, the geometric trajectory misaligns with temporal progression, triggering severe temporal compression. In this paper, we show that embedding norm decay causes Euclidean distance to yield spurious overtaking, indicating that token competition must be evaluated directionally. In this space, a candidate overtakes the source token only when the effective noise magnitude surpasses their semantic separation. We formalize these pairwise dynamics as the Competition Threshold Theory (CTT). Crucially, high-dimensional overtaking requires extreme noise levels, causing standard trajectories to concentrate identity decay into a narrow window near the noisy endpoint. To resolve this, we propose Semantic Retention Scheduling (SRS) to align time allocation with the progression of identity decay, and demonstrate that existing token identity metrics are fundamentally statistical aggregations of these pairwise crossings. Building on these theoretical insights, we introduce LLaCO, a generative framework powered by competition-aware semantic scheduling. LLaCO adapts a pretrained backbone to continuous representations via an embedding directional adapter and timestep-conditioned modulation. To operate within the directional space, LLaCO employs a spherical noising process formulated as a geodesic interpolation toward a mask-biased endpoint, eliminating norm decay while preserving accurate token competition. Guided by SRS, LLaCO reallocates the denoising trajectory proportionally to the logarithmic rank decay. Instantiated as an 8B model, LLaCO outperforms previous continuous diffusion baselines and rivals discrete diffusion models of comparable scale.
Abstract (translated)
URL
https://arxiv.org/abs/2610.32604
PDF
-
PASTA: Prompt-Aware Annealed Diffusion Sampling
2026-10-23 09:16:22Ren Ajiki, Ryugo Morita, Stanislav Frolov, Brian Bernhard Moser, Ko Watanabe, Riku Takahashi, Andreas Dengel, Jinjia ZhouAbstract
While Diffusion models produce high-quality images, real-world visual design requires intentionally preserving clean regions for headlines, logos, or UI elements. Existing layout-aware text-to-image methods struggle to strictly preserve reserved regions, often introducing unwanted content that complicates downstream editing. Existing approaches either manipulate attention to displace content or suppress conditioning via noise-level interventions. However, these methods sacrifice visual coherence, producing flat regions or noticeable style inconsistencies between reserved and non-reserved areas. To address this limitation, we propose a Prompt-Aware Shared-Latent Training-Free Annealed Diffusion Sampling (PASTA), a training-free framework that enforces reserved-region constraints by selectively applying semantic guidance to object-allowed regions during sampling. PASTA operates on a shared latent trajectory, where full-prompt and context-prompt guidance are fused through mask-guided sampling. An annealed schedule suppresses semantic leakage during early denoising stages while gradually restoring visual consistency in later stages. Extensive experiments demonstrate that PASTA improves reserved-region fidelity while preserving overall image realism and perceptual consistency.
Abstract (translated)
URL
https://arxiv.org/abs/2610.32390
PDF
-
Neural Disjunctive Normal Forms for Learning Exact Rank Filters
2026-10-23 05:55:37Diego Marcondes, Wonder A. L. Alves, Ronaldo F. HashimotoAbstract
Gradient-based learning of morphological operators is hindered by locally sparse derivatives with respect to the structuring elements. We develop a method for learning rank filters that retains exact operator evaluation while distributing the derivatives across the window. Using threshold decomposition, we construct a neural disjunctive normal form representation and establish a sufficient condition for exactness with integer-valued images and structuring elements. We then derive a smooth surrogate from the unique root of a smooth threshold equation. Implicit differentiation gives closed-form derivatives with normalised local weights, whose distribution is controlled by a smoothing parameter. These derivatives are used in a straight-through procedure: the forward pass evaluates the exact rank filter, while the backward pass uses the surrogate. For fixed ranks, the procedure learns real-valued structuring elements of any rank filter, including the median, and extends to compositions, infima, and suprema of rank filters. Experiments on image denoising using MNIST and Fashion-MNIST and retinal vessel segmentation show that, starting from a generic random initialisation, the proposed approach consistently improves performance over training with exact subgradients in every setting. In image denoising, our best network also outperforms smooth morphological networks in all settings. On vessel segmentation, the learned erosions recover structuring elements that fit the vessels, as theoretically expected, whereas most of those trained with exact subgradients remain close to their initialisation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.32203
PDF
-
DELA: Learning Controllable Edits for Diffusion Language Models
2026-10-23 02:57:27Xinsong Feng, Chenan Wang, Leshu Tang, Johnny Liu, Dong Li, Da Li, Haipeng ChenAbstract
Diffusion language models (DLMs) support bidirectional denoising and iterative refinement, but generation remains largely controlled by token-level masking schedules. This leaves draft-level refinement implicit and limits explicit control over when to insert missing content, delete redundant spans, revisit uncertain regions, or stop editing. We introduce Diffusion Editing Language Adapter (DELA), a plug-in framework that enables controllable draft refinement with a frozen DLM. DELA operates on clean, variable-length drafts using four explicit editing actions: Insert, Delete, Remask, and Stop. The controller determines when and where to edit the draft, while the frozen DLM fills masked regions through denoising, separating high-level edit control from token-level generation. A naive single-phase editor would be computationally expensive because arbitrary interior edits can disrupt prefix stability and limit reliable KV-cache reuse. To address this, DELA uses a two-phase decoding procedure: a fast, append-only drafting phase that supports KV-cache reuse, followed by a global refinement phase that enables targeted non-monotonic edits to the full draft. We train the controller using a staged curriculum that begins with learning local edit priors from synthetic recovery tasks and mined actions that improve draft quality. Subsequent stages refine edit selection through rollout-based outcome preferences and alignment with downstream task feedback. Experiments on mathematical reasoning and code generation show that DELA achieves gains of up to 5.5 percentage points over direct diffusion decoding while keeping the backbone frozen. DELA thus provides a modular approach to explicit draft-level control in diffusion language models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.32096
PDF
