In a surge of text-to-image (T2I) models and their customization methods that generate new images of a user-provided subject, current works focus on alleviating the costs incurred by a lengthy per-subject optimization. These zero-shot customization methods encode the image of a specified subject into a visual embedding which is then utilized alongside the textual embedding for diffusion guidance. The visual embedding incorporates intrinsic information about the subject, while the textual embedding provides a new context. However, the existing methods often 1) generate images with the same pose as an input image, and 2) exhibit deterioration in the subject's identity when facing a pose variation prompt. We first pin down the problem and show that redundant pose information in the visual embedding interferes with the pose indication in the textual embedding. Conversely, the textual embedding also harms the subject's identity which is tightly entangled with the pose in the visual embedding. As a remedy, we propose text-orthogonal visual embedding which effectively harmonizes with the given textual embedding. We also adopt the visual-only embedding and inject the subject's clear features utilizing a self-attention swap. Our method is both effective and robust, offering highly flexible zero-shot generation while effectively maintaining the subject's identity.
论文检索
输入标题、作者或关键词，从 9,256 篇学术成果中精准定位
Diffusion Transformers have emerged as the preeminent models for a wide array of generative tasks, demonstrating superior performance and efficacy across various applications. The promising results come at the cost of slow inference, as each denoising step requires running the whole transformer model with a large amount of parameters. In this paper, we show that performing the full computation of the model at each diffusion step is unnecessary, as some computations can be skipped by lazily reusing the results of previous steps. Furthermore, we show that the lower bound of similarity between outputs at consecutive steps is notably high, and this similarity can be linearly approximated using the inputs. To verify our demonstrations, we propose the **LazyDiT**, a lazy learning framework that efficiently leverages cached results from earlier steps to skip redundant computations. Specifically, we incorporate lazy learning layers into the model, effectively trained to maximize laziness, enabling dynamic skipping of redundant computations. Experimental results show that LazyDiT outperforms the DDIM sampler across multiple diffusion transformer models at various resolutions. Furthermore, we implement our method on mobile devices, achieving better performance than DDIM with similar latency.
Score-based diffusion models have emerged as effective approaches for both conditional and unconditional generation. Still conditional generation is based on either a specific training of a conditional model or classifier guidance, which requires training a noise-dependent classifier, even when a classifier for uncorrupted data is given. We propose a method that, given a pre-trained unconditional score-based generative model, samples from the conditional distribution under arbitrary logical constraints, without requiring additional training. Differently from other zero-shot techniques, that rather aim at generating valid conditional samples, our method is designed for approximating the true conditional distribution. Firstly, we show how to manipulate the learned score in order to sample from an un-normalized distribution conditional on a user-defined constraint. Then, we define a flexible and numerically stable neuro-symbolic framework for encoding soft logical constraints. Combining these two ingredients we obtain a general, but approximate, conditional sampling algorithm. We further developed effective heuristics aimed at improving the approximation. Finally, we show the effectiveness of our approach in approximating conditional distributions for various types of constraints and data: tabular data, images and time series.
Diffusion-based molecular graph generative models have achieved significant success in template-free, single-step retrosynthesis prediction. However, these models typically generate reactants from scratch, often overlooking the fact that the scaffold of a product molecule typically remains unchanged during chemical reactions. To leverage this useful observation, we introduce a retrieval-augmented molecular graph generation framework. Our framework comprises three key components: a retrieval component that identifies similar molecules for the given product, an integration component that learns valuable clues from these molecules about which part of the product should remain unchanged, and a base generative model that is prompted by these clues to generate the corresponding reactants. We explore various design choices for critical and under-explored aspects of this framework and instantiate it as the Retrieval-Augmented RetroBridge (RARB). RARB demonstrates state-of-the-art performance on standard benchmarks, achieving a 14.8% relative improvement in top-1 accuracy over its base generative model, highlighting the effectiveness of retrieval augmentation. Additionally, RARB excels in handling out-of-distribution molecules, and its advantages remain significant even with smaller models or fewer denoising steps. These strengths make RARB highly valuable for real-world retrosynthesis applications, where extrapolation to novel molecules and high-throughput prediction are essential.
Denoising Diffusion Variational Inference: Diffusion Models as Expressive Variational Posteriors
PDF ↗We propose denoising diffusion variational inference (DDVI), a black-box variational inference algorithm for latent variable models which relies on diffusion models as flexible approximate posteriors. Specifically, our method introduces an expressive class of diffusion-based variational posteriors that perform iterative refinement in latent space; we train these posteriors with a novel regularized evidence lower bound (ELBO) on the marginal likelihood inspired by the wake-sleep algorithm. Our method is easy to implement (it fits a regularized extension of the ELBO), is compatible with black-box variational inference, and outperforms alternative classes of approximate posteriors based on normalizing flows or adversarial networks. We find that DDVI improves inference and learning in deep latent variable models across common benchmarks as well as on a motivating task in biology---inferring latent ancestry from human genomes---where it outperforms strong baselines on the Thousand Genomes dataset.
Incomplete multi-view multi-label classification aims to accurately predict labels for each sample in the face of some missing views. Due to its widespread presence in real-world scenarios, it has become an extensively researched topic. In addition to the challenges brought by missing views, it also encounters issues caused by redundant views, whose inclusion fails to make a positive contribution to performance. In this paper, we make the first attempt to take advantage of diffusion models to address the missing view problem and design a strategy to identify and remove redundant views. Specifically, we train a diffusion model conditioned on the pseudo-labels to recover information of missing views. The learned diffusion model can carry data distribution knowledge in training split to the data. Regarding redundant identification strategy, it is designed by considering both the additional information of views and the classification difficulty level of samples, thereby adaptively identifying and removing redundant views. We conduct extensive experiments on five datasets, and the proposed method achieves favorable performance against several state-of-the-art methods on the multi-view multi-label classification task.
Learning graph generative models over latent spaces has received less attention compared to models that operate on the original data space and has so far demonstrated lacklustre performance. We present GLAD a latent space graph generative model. Unlike most previous latent space graph generative models, GLAD operates on a discrete latent space that preserves to a significant extent the discrete nature of the graph structures making no unnatural assumptions such as latent space continuity. We learn the prior of our discrete latent space by adapting diffusion bridges to its structure. By operating over an appropriately constructed latent space we avoid relying on decompositions that are often used in models that operate in the original data space. We present experiments on a series of graph benchmark datasets that demonstrates GLAD as the first equivariant latent graph generative method achieves competitive performance with the state of the art baselines.
In this paper we describe a novel framework for diffusion-based generative modeling on constrained spaces. In particular, we introduce manual bridges, a framework that expands the kinds of constraints that can be practically used to form so-called diffusion bridges. We develop a mechanism for combining multiple such constraints so that the resulting multiply-constrained model remains a manual bridge that respects all constraints. We also develop a mechanism for training a diffusion model that respects such multiple constraints while also adapting it to match a data distribution. We develop and extend theory demonstrating the mathematical validity of our mechanisms. Additionally, we demonstrate our mechanism in constrained generative modeling tasks, highlighting a particular high-value application in modeling trajectory initializations for path planning and control in autonomous vehicles.
Recent advancements in sequence modeling have led to the development of the Mamba architecture, noted for its selective state space approach, offering a promising avenue for efficient long sequence handling. However, its application in 3D shape generation, particularly at high resolutions, remains underexplored. Traditional diffusion transformers (DiT) with self-attention mechanisms, despite their potential, face scalability challenges due to the cubic complexity of attention operations as input length increases. This complexity becomes a significant hurdle when dealing with high-resolution voxel sizes. To address this challenge, we introduce a novel diffusion architecture tailored for 3D point clouds generation—Diffusion Mamba (DiM-3D). This architecture forgoes traditional attention mechanisms, instead utilizing the inherent efficiency of the Mamba architecture to maintain linear complexity with respect to sequence length. DiM-3D is characterized by fast inference times and substantially lower computational demands, quantified in reduced Gflops, thereby addressing the key scalability issues of prior models. Our empirical results on the ShapeNet benchmark demonstrate that DiM-3D achieves state-of-the-art performance in generating high-fidelity and diverse 3D shapes. Additionally, DiM-3D shows superior capabilities in tasks like 3D point cloud completion. This not only proves the model's scalability but also underscores its efficiency in generating detailed, high-resolution voxels necessary for advanced 3D shape modeling, particularly excelling in environments requiring 256-resolution voxel sizes. Through these findings, we illustrate the exceptional scalability and efficiency of the Diffusion Mamba framework in 3D shape generation, setting a new standard for the field and paving the way for future explorations in high-resolution 3D modeling technologies.
VVRec: Reconstruction Attacks on DL-based Volumetric Video Upstreaming via Latent Diffusion Model with Gamma Distribution
PDF ↗With the popularity of 3D volumetric video applications, such as Autonomous Driving, Virtual Reality, and Mixed Reality, current developers have turned to deep learning for compressing volumetric video frames, i.e., point clouds for video upstreaming. The latest deep learning-based solutions offer higher efficiency, lower distortion, and better hardware support compared to traditional ones like MPEG and JPEG. However, privacy threats arise, especially reconstruction attacks targeting to recover the original input point cloud from the intermediate results. In this paper, we design VVRec, to the best of our knowledge, which is the first targeting DL-based Volumetric Video Reconstruction attack scheme. VVRec demonstrates the ability to reconstruct high-quality point clouds from intercepted transmission intermediate results using four well-trained neural network modules we design. Leveraging the latest latent diffusion models with Gamma distribution and a refinement algorithm, VVRec excels in reconstruction quality and color recovery and surpasses existing defenses. We evaluate VVRec using three volumetric video datasets. The results demonstrate that VVRec achieves 64.70dB reconstruction accuracy, with an impressive 46.39% reduction of distortion over baselines.
Decentralized federated learning (DFL) realizes cooperative model training among connected clients without relying on a central server, thereby mitigating communication bottlenecks and eliminating the single-point failure issue present in centralized federated learning (CFL). Most existing work on DFL focuses on supervised learning, assuming each client possesses sufficient labeled data for local training. However, in real-world applications, much of the data is unlabeled. We address this by considering a challenging yet practical semi-supervised learning (SSL) scenario in DFL, where clients may have varying data sources: some with few labeled samples, some with purely unlabeled data, and others with both. In this work, we propose SemiDFL, the first semi-supervised DFL method that enhances DFL performance in SSL scenarios by establishing a consensus in both data and model spaces. Specifically, we utilize neighborhood information to improve the quality of pseudo-labeling, which is crucial for effectively leveraging unlabelled data. We then design a consensus-based diffusion model to generate synthesized data, which is used in combination with pseudo-labeled data to create mixed datasets. Additionally, we develop an adaptive aggregation method that leverages the model accuracy of synthesized data to further enhance SemiDFL performance. Through extensive experimentation, we demonstrate the remarkable performance superiority of the proposed DFL-Semi method over existing CFL and DFL schemes in both iid and Non-iid SSL scenarios.
Although diffusion models have achieved remarkable success in the field of image generation, their latent space remains under-explored. Current methods for identifying semantics within latent space often rely on external supervision, such as textual information and segmentation masks. In this paper, we propose a method to identify semantic attributes in the latent space of pre-trained diffusion models without any further training. By projecting the Jacobian of the targeted semantic region into a low-dimensional subspace which is orthogonal to the non-masked regions, our approach facilitates precise semantic discovery and control over local masked areas, eliminating the need for annotations. We conducted extensive experiments across multiple datasets and various architectures of diffusion models, achieving state-of-the-art performance. In particular, for some specific face attributes, the performance of our proposed method even surpasses that of supervised approaches, demonstrating its superior ability in editing local image properties.
In this paper, we present a novel diffusion model-based monaural speech enhancement method. Our approach incorporates the separate estimation of speech spectra's magnitude and phase in two diffusion networks. Throughout the diffusion process, noise clips from real-world noise interferences are added gradually to the clean speech spectra and a noise-aware reverse process is proposed to learn how to generate both clean speech spectra and noise spectra. Furthermore, to fully leverage the intrinsic relationship between magnitude and phase, we introduce a complex-cycle-consistent (CCC) mechanism that uses the estimated magnitude to map the phase, and vice versa. We implement this algorithm within a phase-aware speech enhancement diffusion model (SEDM). We conduct extensive experiments on public datasets to demonstrate the effectiveness of our method, highlighting the significant benefits of exploiting the intrinsic relationship between phase and magnitude information to enhance speech. The comparison to conventional diffusion models demonstrates the superiority of SEDM.
Many real-world data are inherently multi-dimensional, e.g., color images, videos, and hyperspectral images. How to effectively and compactly represent these multi-dimensional data within a unified framework is an important pursuit. Previous methods focus on tensor factorizations, convolutional networks, or diffusion models for multi-dimensional data representation, which may not fully utilize inherent data structures and may lead to redundant parameters. In this work, we propose a Deep Rank-One Tensor Functional Factorization (DRO-TFF), which internally utilizes more comprehensive data priors facilitated by much fewer parameters. Concretely, our DRO-TFF consists of three organically integrated blocks: compact rank-one factorizations in the spatial domain, a deep transform to capture underlying low-dimensional structures, and smooth factors parameterized by implicit neural representations. Through a series of theoretical analysis, we show the rich data priors encoded in the DRO-TFF structure, e.g., Lipschitz smoothness and low-rankness. Extensive experiments on multi-dimensional data recovery problems, such as image and video inpainting, image denoising, and hyperspectral mixed noise removal, showcase the effectiveness of the proposed method.
Diffusion models (DMs) have attracted attention in generative modeling due to their ability to produce high-quality, diverse outputs by progressively adding noise to data and then denoising it. However, DMs are computationally intensive due to their iterative nature, requiring numerous forward passes and high-precision operations, making them less efficient for resource-constrained environments. Recent efforts to reduce these computational demands using quantization show promise by converting high-precision parameters to lower precision, but they face challenges unique to DMs, particularly in addressing cross-timestep error propagation in the iterative process. In this paper, we analyze cross-timestep error propagation in quantized DMs, revealing that previous methods focusing only on reducing noise estimation discrepancies are insufficient. Instead, we introduce Cross-Timestep Error Correction (CTEC), where the quantized model not only approximates the full-precision model but also corrects errors from the previous timestep. A distillation method is applied to learn this correction process effectively. We conduct extensive experiments on unconditional image generation with LSUN-Churches and LSUN-Bedrooms, as well as conditional image generation with ImageNet. Our findings demonstrate the effectiveness of our method in significantly reducing accumulated quantization errors across timesteps within the quantized diffusion process. This enhancement enables the generation of high-quality images, even when constrained by reduced bitwidths.
Diffusion models have shown promising ability in generating high-quality time series (TS) data. Despite the initial success, existing works mostly focus on the authenticity of data at the individual level, but pay less attention to preserving the population-level properties on the entire dataset. Such population-level properties include value distributions for each dimension and distributions of certain functional dependencies (e.g., cross-correlation, CC) between different dimensions. For instance, when generating house energy consumption TS data, the value distributions of the outside temperature and the kitchen temperature should be preserved, as well as the distribution of CC between them. Preserving such TS population-level properties is critical in maintaining the statistical insights of the datasets, mitigating model bias, and augmenting downstream tasks like TS prediction. Yet, it is often overlooked by existing models. Hence, data generated by existing models often bear distribution shifts from the original data. We propose Population-aware Diffusion for Time Series (PaD-TS), a new TS generation model that better preserves the population-level properties. The key novelties of PaD-TS include 1) a new training method explicitly incorporating TS population-level property preservation, and 2) a new dual-channel encoder model architecture that better captures the TS data structure. Empirical results in major benchmark datasets show that PaD-TS can improve the average CC distribution shift score between real and synthetic data by 5.9x while maintaining a performance comparable to state-of-the-art models on individual-level authenticity.
CG-TGAN: Conditional Generative Adversarial Networks with Graph Neural Networks for Tabular Data Synthesizing
PDF ↗Data sharing is necessary for AI to be widely used, but sharing sensitive data with others with privacy is risky. To solve these problems, it is necessary to synthesize realistic tabular data. In many cases, tabular data contains a mixture of continuous, mixed, categorical columns. Moreover, columns of the same type may have multimodal distribution or be highly imbalanced. These issues make it challenging to synthesize tabular data. The synthesized tabular data should reflect the relational meaning between columns of tabular data, so modeling the probability distribution of tabular data is a non-trivial task. Traditional tabular data synthesizing models are based on GAN or diffusion models and are built using fully connected or convolutional layers. However, fully connected layers have the disadvantage of low inductive bias, and convolutional layers are not invariant to the column order of tabular data. Therefore, we assume that converting tabular data into graph-structured data and using a graph neural network would produce better synthetic data than using fully connected layers or convolutional layers. Our study aims to show that GANs constructed with graph neural networks can outperform existing GAN models using fully connected layers or convolutional layers. We propose CG-TGAN, a conditional GAN built using graph neural networks. To learn how to synthesize realistic data, the graph neural networks in the discriminator and generator learn graph-level tasks and node-level tasks together. The discriminator of CG-TGAN learns a graph-level task to distinguish between real and synthetic data and node-level tasks to predict the value of the target node. CG-TGAN’s generator learns a graph-level task to synthesize an overall graph similar to real data and node-level tasks to learn how to synthesize a fake graph with the proper relation between nodes. In this paper, we show that CG-TGAN outperforms GAN-based models and is comparable to diffusion-based models.
Federated Learning (FL) offers a decentralized approach to model training, where data remains local and only model parameters are shared between the clients and the central server. Traditional methods, such as Federated Averaging (FedAvg), linearly aggregate these parameters which are usually trained on heterogeneous data distributions, potentially overlooking the complex, high-dimensional nature of the parameter space. This can result in degraded performance of the aggregated model. While personalized FL approaches can mitigate the heterogeneous data issue to some extent, the limitation of linear aggregation remains unresolved. To alleviate this issue, we investigate the generative approach of diffusion model and propose a novel generative parameter aggregation framework for personalized FL, pFedGPA. In this framework, we deploy a diffusion model on the server to integrate the diverse parameter distributions and propose a parameter inversion method to efficiently generate a set of personalized parameters for each client. This inversion method transforms the uploaded parameters into a latent code, which is then aggregated through denoising sampling to produce the final personalized parameters. By encoding the dependence of a client's model parameters on the specific data distribution using the high-capacity diffusion model, pFedGPA can effectively decouple the complexity of the overall distribution of all clients' model parameters from the complexity of each individual client's parameter distribution. Our experimental results consistently demonstrate the superior performance of the proposed method across multiple datasets, surpassing baseline approaches.
In privacy-preserving distributed learning environments, data stored on local clients cannot be shared with other clients or servers. We consider a new active learning problem setup for these environments, where the server aims to build a centralized model by distributing labeling budgets across different clients. Our algorithm identifies which clients and their data points warrant annotation by estimating the global impact of the resulting labels. We evaluate this impact by embedding the clients into the manifold of learner parameters, formed by the task learner's predictions on unlabeled data, and diffusing the reduction in predictive uncertainties caused by labeling. The algorithm effectively selects clients with high estimated impact while achieving diversity in client selection, all without accessing local client data. In experiments, our approach demonstrates substantial advancements when compared to adaptations of existing active learning algorithms.
While pruning methods effectively maintain model performance without extra training costs, they often focus solely on preserving crucial connections, overlooking the impact of pruned weights on subsequent fine-tuning or distillation, leading to inefficiencies. Moreover, most compression techniques for generative models have been developed primarily for GANs, tailored to specific architectures like StyleGAN, and research into compressing Diffusion models has just begun. Even more, these methods are often applicable only to GANs or Diffusion models, highlighting the need for approaches that work across both model types. In this paper, we introduce Singular Value Scaling (SVS), a versatile technique for refining pruned weights, applicable to both model types. Our analysis reveals that pruned weights often exhibit dominant singular vectors, hindering fine-tuning efficiency and leading to suboptimal performance compared to random initialization. Our method enhances weight initialization by minimizing the disparities between singular values of pruned weights, thereby improving the fine-tuning process. This approach not only guides the compressed model toward superior solutions but also significantly speeds up fine-tuning. Extensive experiments on StyleGAN2, StyleGAN3 and DDPM demonstrate that SVS improves compression performance across model types without additional training costs.