NeurIPS 2023 论文列表

Deep generative models can generate high-fidelity audio conditioned on various types of representations (e.g., mel-spectrograms, Mel-frequency Cepstral Coefficients (MFCC)). Recently, such models have been used to synthesize audio waveforms conditioned on highly compressed representations. Although such methods produce impressive results, they are prone to generate audible artifacts when the conditioning is flawed or imperfect. An alternative modeling approach is to use diffusion models. However, these have mainly been used as speech vocoders (i.e., conditioned on mel-spectrograms) or generating relatively low sampling rate signals. In this work, we propose a high-fidelity multi-band diffusion-based framework that generates any type of audio modality (e.g., speech, music, environmental sounds) from low-bitrate discrete representations. At equal bit rate, the proposed approach outperforms state-of-the-art generative techniques in terms of perceptual quality. Training and evaluation code are available on the facebookresearch/ audiocraft github project. Samples are available on the following link (https://ai.honu.io/papers/mbd/).

The past few years have witnessed great success in the use of diffusion models (DMs) to generate high-fidelity images with the help of stochastic differential equations (SDEs). However, discretization error is an inevitable limitation when utilizing numerical solvers to solve SDEs. To address this limitation, we provide a theoretical analysis demonstrating that an appropriate combination of the contrastive loss and score matching serves as an upper bound of the KL divergence between the true data distribution and the model distribution. To obtain this bound, we utilize a contrastive loss to construct a contrastive sampling chain to fine-tuning the pre-trained DM. In this manner, our method reduces the discretization error and thus yields a smaller gap between the true data distribution and our model distribution. Moreover, the presented method can be applied to fine-tuning various pre-trained DMs, both with or without fast sampling algorithms, contributing to better sample quality or slightly faster sampling speeds. To validate the efficacy of our method, we conduct comprehensive experiments. For example, on CIFAR10, when applied to a pre-trained EDM, our method improves the FID from 2.04 to 1.88 with 35 neural function evaluations (NFEs), and reduces NFEs from 35 to 25 to achieve the same 2.04 FID.

Generative modeling has recently undergone remarkable advancements, primarily propelled by the transformative implications of Diffusion Probabilistic Models (DPMs). The impressive capability of these models, however, often entails significant computational overhead during both training and inference. To tackle this challenge, we present Diff-Pruning, an efficient compression method tailored for learning lightweight diffusion models from pre-existing ones, without the need for extensive re-training. The essence of Diff-Pruning is encapsulated in a Taylor expansion over pruned timesteps, a process that disregards non-contributory diffusion steps and ensembles informative gradients to identify important weights. Our empirical assessment, undertaken across several datasets highlights two primary benefits of our proposed method: 1) Efficiency: it enables approximately a 50\% reduction in FLOPs at a mere 10% to 20% of the original training expenditure; 2) Consistency: the pruned diffusion models inherently preserve generative behavior congruent with their pre-trained models.

Generative models produce unbounded outputs, necessitating the use of refusal techniques to confine their output space. Employing generative refusals is crucial in upholding the ethical and copyright integrity of synthesized content, particularly when working with widely adopted diffusion models. "Concept negation'' presents a promising paradigm to achieve generative refusals, as it effectively defines and governs the model's output space based on concepts, utilizing natural language interfaces that are readily comprehensible to humans. However, despite the valuable contributions of prior research to the field of concept negation, it still suffers from significant limitations. The existing concept negation methods, which operate based on the composition of score or noise predictions from the diffusion process, are limited to independent concepts (e.g., ``a blonde girl`` without ``glasses``) and fail to consider the interconnected nature of concepts in reality (e.g., ``Mickey mouse eats ice cream`` without ``Disney characters``). Keeping the limitations in mind, we propose a novel framework, called $ProtoRe$, to improve the flexibility of concept negation via test-time negative concept identification along with purification in the feature space. $ProtoRe$ works by incorporating CLIP's language-contrastive knowledge to identify the prototype of negative concepts, extract the negative features from outputs using the prototype as a prompt, and further refine the attention maps by retrieving negative features. Our evaluation on multiple benchmarks shows that $ProtoRe$ outperforms state-of-the-art methods under various settings, in terms of the effectiveness of purification and the fidelity of generative images.

Recent progress in music generation has been remarkably advanced by the state-of-the-art MusicLM, which comprises a hierarchy of three LMs, respectively, for semantic, coarse acoustic, and fine acoustic modelings. Yet, sampling with the MusicLM requires processing through these LMs one by one to obtain the fine-grained acoustic tokens, making it computationally expensive and prohibitive for a real-time generation. Efficient music generation with a quality on par with MusicLM remains a significant challenge. In this paper, we present **M**e**L**o**D**y (**M** for music; **L** for LM; **D** for diffusion), an LM-guided diffusion model that generates music audios of state-of-the-art quality meanwhile reducing 95.7\% to 99.6\% forward passes in MusicLM, respectively, for sampling 10s to 30s music. MeLoDy inherits the highest-level LM from MusicLM for semantic modeling, and applies a novel dual-path diffusion (DPD) model and an audio VAE-GAN to efficiently decode the conditioning semantic tokens into waveform. DPD is proposed to simultaneously model the coarse and fine acoustics by incorporating the semantic information into segments of latents effectively via cross-attention at each denoising step. Our experimental results suggest the superiority of MeLoDy, not only in its practical advantages on sampling speed and infinitely continuable generation, but also in its state-of-the-art musicality, audio quality, and text correlation. Our samples are available at https://Efficient-MeLoDy.github.io/.

Vision-and-Language Navigation requires the agent to follow language instructions to navigate through 3D environments. One main challenge in Vision-and-Language Navigation is the limited availability of photorealistic training environments, which makes it hard to generalize to new and unseen environments. To address this problem, we propose PanoGen, a generation method that can potentially create an infinite number of diverse panoramic environments conditioned on text. Specifically, we collect room descriptions by captioning the room images in existing Matterport3D environments, and leverage a state-of-the-art text-to-image diffusion model to generate the new panoramic environments. We use recursive outpainting over the generated images to create consistent 360-degree panorama views. Our new panoramic environments share similar semantic information with the original environments by conditioning on text descriptions, which ensures the co-occurrence of objects in the panorama follows human intuition, and creates enough diversity in room appearance and layout with image outpainting. Lastly, we explore two ways of utilizing PanoGen in VLN pre-training and fine-tuning. We generate instructions for paths in our PanoGen environments with a speaker built on a pre-trained vision-and-language model for VLN pre-training, and augment the visual observation with our panoramic environments during agents' fine-tuning to avoid overfitting to seen environments. Empirically, learning with our PanoGen environments achieves the new state-of-the-art on the Room-to-Room, Room-for-Room, and CVDN datasets. Besides, we find that pre-training with our PanoGen speaker data is especially effective for CVDN, which has under-specified instructions and needs commonsense knowledge to reach the target. Lastly, we show that the agent can benefit from training with more generated panoramic environments, suggesting promising results for scaling up the PanoGen environments to enhance agents' generalization to unseen environments.

We present a novel algorithm for text-driven image-to-image translation based on a pretrained text-to-image diffusion model. Our method aims to generate a target image by selectively editing regions of interest in a source image, defined by a modifying text, while preserving the remaining parts. In contrast to existing techniques that solely rely on a target prompt, we introduce a new score function that additionally considers both the source image and the source text prompt, tailored to address specific translation tasks. To this end, we derive the conditional score function in a principled way, decomposing it into the standard score and a guiding term for target image generation. For the gradient computation about the guiding term, we assume a Gaussian distribution for the posterior distribution and estimate its mean and variance to adjust the gradient without additional training. In addition, to improve the quality of the conditional score guidance, we incorporate a simple yet effective mixup technique, which combines two cross-attention maps derived from the source and target latents. This strategy is effective for promoting a desirable fusion of the invariant parts in the source image and the edited regions aligned with the target prompt, leading to high-fidelity target image generation. Through comprehensive experiments, we demonstrate that our approach achieves outstanding image-to-image translation performance on various tasks. Code is available at https://github.com/Hleephilip/CSG.

Machine learning models are increasingly being scaled in both sequence length and model dimension to reach longer contexts and better performance. However, existing architectures such as Transformers scale quadratically along both these axes. We ask: are there performant architectures that can scale sub-quadratically along sequence length and model dimension? We introduce Monarch Mixer (M2), a new architecture that uses the same sub-quadratic primitive along both sequence length and model dimension: Monarch matrices, a simple class of expressive structured matrices that captures many linear transforms, achieves high hardware efficiency on GPUs, and scales sub-quadratically. As a proof of concept, we explore the performance of M2 in three domains: non-causal BERT-style language modeling, ViT-style image classification, and causal GPT-style language modeling. For non-causal BERT-style modeling, M2 matches BERT-base and BERT-large in downstream GLUE quality with up to 27% fewer parameters, and achieves up to 9.1$\times$ higher throughput at sequence length 4K. On ImageNet, M2 outperforms ViT-b by 1% in accuracy, with only half the parameters. Causal GPT-style models introduce a technical challenge: enforcing causality via masking introduces a quadratic bottleneck. To alleviate this bottleneck, we develop a novel theoretical view of Monarch matrices based on multivariate polynomial evaluation and interpolation, which lets us parameterize M2 to be causal while remaining sub-quadratic. Using this parameterization, M2 matches GPT-style Transformers at 360M parameters in pretraining perplexity on The PILE—showing for the first time that it may be possible to match Transformer quality without attention or MLPs.

This paper studies a diffusion-based framework to address the low-light image enhancement problem. To harness the capabilities of diffusion models, we delve into this intricate process and advocate for the regularization of its inherent ODE-trajectory. To be specific, inspired by the recent research that low curvature ODE-trajectory results in a stable and effective diffusion process, we formulate a curvature regularization term anchored in the intrinsic non-local structures of image data, i.e., global structure-aware regularization, which gradually facilitates the preservation of complicated details and the augmentation of contrast during the diffusion process. This incorporation mitigates the adverse effects of noise and artifacts resulting from the diffusion process, leading to a more precise and flexible enhancement. To additionally promote learning in challenging regions, we introduce an uncertainty-guided regularization technique, which wisely relaxes constraints on the most extreme regions of the image. Experimental evaluations reveal that the proposed diffusion-based framework, complemented by rank-informed regularization, attains distinguished performance in low-light enhancement. The outcomes indicate substantial advancements in image quality, noise suppression, and contrast amplification in comparison with state-of-the-art methods. We believe this innovative approach will stimulate further exploration and advancement in low-light image processing, with potential implications for other applications of diffusion models. The code is publicly available at https://github.com/jinnh/GSAD.

Diffusion models are powerful generative models but suffer from slow sampling, often taking 1000 sequential denoising steps for one sample. As a result, considerable efforts have been directed toward reducing the number of denoising steps, but these methods hurt sample quality. Instead of reducing the number of denoising steps (trading quality for speed), in this paper we explore an orthogonal approach: can we run the denoising steps in parallel (trading compute for speed)? In spite of the sequential nature of the denoising steps, we show that surprisingly it is possible to parallelize sampling via Picard iterations, by guessing the solution of future denoising steps and iteratively refining until convergence. With this insight, we present ParaDiGMS, a novel method to accelerate the sampling of pretrained diffusion models by denoising multiple steps in parallel. ParaDiGMS is the first diffusion sampling method that enables trading compute for speed and is even compatible with existing fast sampling techniques such as DDIM and DPMSolver. Using ParaDiGMS, we improve sampling speed by 2-4x across a range of robotics and image generation models, giving state-of-the-art sampling speeds of 0.2s on 100-step DiffusionPolicy and 14.6s on 1000-step StableDiffusion-v2 with no measurable degradation of task reward, FID score, or CLIP score.

Embodied agents operate in a structured world, often solving tasks with spatial, temporal, and permutation symmetries. Most algorithms for planning and model-based reinforcement learning (MBRL) do not take this rich geometric structure into account, leading to sample inefficiency and poor generalization. We introduce the Equivariant Diffuser for Generating Interactions (EDGI), an algorithm for MBRL and planning that is equivariant with respect to the product of the spatial symmetry group SE(3), the discrete-time translation group ℤ, and the object permutation group Sₙ. EDGI follows the Diffuser framework by Janner et al. (2022) in treating both learning a world model and planning in it as a conditional generative modeling problem, training a diffusion model on an offline trajectory dataset. We introduce a new SE(3) × ℤ × Sₙ-equivariant diffusion model that supports multiple representations. We integrate this model in a planning loop, where conditioning and classifier guidance let us softly break the symmetry for specific tasks as needed. On object manipulation and navigation tasks, EDGI is substantially more sample efficient and generalizes better across the symmetry group than non-equivariant models.

Diffusion models excel at producing high-quality samples but naively require hundreds of iterations, prompting multiple attempts to distill the generation process into a faster network. However, many existing approaches suffer from a variety of challenges: the process for distillation training can be complex, often requiring multiple training stages, and the resulting models perform poorly when utilized in single-step generative applications. In this paper, we introduce a simple yet effective means of distilling diffusion models *directly* from the initial noise to the resulting image. Of particular importance to our approach is to leverage a new Deep Equilibrium (DEQ) model as the distilled architecture: the Generative Equilibrium Transformer (GET). Our method enables fully offline training with just noise/image pairs from the diffusion model while achieving superior performance compared to existing one-step methods on comparable training budgets. We demonstrate that the DEQ architecture is crucial to this capability, as GET matches a $5\times$ larger ViT in terms of FID scores while striking a critical balance of computational cost and image quality. Code, checkpoints, and datasets are available [here](https://github.com/locuslab/get).

Recent works have shown that diffusion models can learn essentially any distribution provided one can perform score estimation. Yet it remains poorly understood under what settings score estimation is possible, let alone when practical gradient-based algorithms for this task can provably succeed. In this work, we give the first provably efficient results for one of the most fundamental distribution families, Gaussian mixture models. We prove that GD on the denoising diffusion probabilistic model (DDPM) objective can efficiently recover the ground truth parameters of the mixture model in the following two settings: 1. We show GD with random initialization learns mixtures of two spherical Gaussians in $d$ dimensions with $1/\text{poly}(d)$-separated centers. 2. We show GD with a warm start learns mixtures of $K$ spherical Gaussians with $\Omega(\sqrt{\log(\min(K,d))})$-separated centers. A key ingredient in our proofs is a new connection between score-based methods and two other approaches to distribution learning, EM and spectral methods.

Many unsupervised methods have recently been proposed for multivariate time series anomaly detection. However, existing works mainly focus on stable data yet often omit the drift generated from non-stationary environments, which may lead to numerous false alarms. We propose **D**ynamic **D**ecomposition with **D**iffusion **R**econstruction (D$^3$R), a novel anomaly detection network for real-world unstable data to fill the gap. D$^3$R tackles the drift via decomposition and reconstruction. In the decomposition procedure, we utilize data-time mix-attention to dynamically decompose long-period multivariate time series, overcoming the limitation of the local sliding window. The information bottleneck is critical yet difficult to determine in the reconstruction procedure. To avoid retraining once the bottleneck changes, we control it externally by noise diffusion and directly reconstruct the polluted data. The whole model can be trained end-to-end. Extensive experiments on various real-world datasets demonstrate that D$^3$R significantly outperforms existing methods, with a 11% average relative improvement over the previous SOTA models.

We propose a new score-based approach to generate 3D molecules represented as atomic densities on regular grids. First, we train a denoising neural network that learns to map from a smooth distribution of noisy molecules to the distribution of real molecules. Then, we follow the _neural empirical Bayes_ framework [Saremi and Hyvarinen, 2019] and generate molecules in two steps: (i) sample noisy density grids from a smooth distribution via underdamped Langevin Markov chain Monte Carlo, and (ii) recover the "clean" molecule by denoising the noisy grid with a single step. Our method, _VoxMol_, generates molecules in a fundamentally different way than the current state of the art (ie, diffusion models applied to atom point clouds). It differs in terms of the data representation, the noise model, the network architecture and the generative modeling algorithm. Our experiments show that VoxMol captures the distribution of drug-like molecules better than state of the art, while being faster to generate samples.

Text-guided image editing is widely needed in daily life, ranging from personal use to professional applications such as Photoshop. However, existing methods are either zero-shot or trained on an automatically synthesized dataset, which contains a high volume of noise. Thus, they still require lots of manual tuning to produce desirable outcomes in practice. To address this issue, we introduce MagicBrush, the first large-scale, manually annotated dataset for instruction-guided real image editing that covers diverse scenarios: single-turn, multi-turn, mask-provided, and mask-free editing. MagicBrush comprises over 10K manually annotated triplets (source image, instruction, target image), which supports trainining large-scale text-guided image editing models. We fine-tune InstructPix2Pix on MagicBrush and show that the new model can produce much better images according to human evaluation. We further conduct extensive experiments to evaluate current image editing baselines from multiple dimensions including quantitative, qualitative, and human evaluations. The results reveal the challenging nature of our dataset and the gap between current baselines and real-world editing needs.

In this paper, we introduce an SE(3) diffusion model-based point cloud registration framework for 6D object pose estimation in real-world scenarios. Our approach formulates the 3D registration task as a denoising diffusion process, which progressively refines the pose of the source point cloud to obtain a precise alignment with the model point cloud. Training our framework involves two operations: An SE(3) diffusion process and an SE(3) reverse process. The SE(3) diffusion process gradually perturbs the optimal rigid transformation of a pair of point clouds by continuously injecting noise (perturbation transformation). By contrast, the SE(3) reverse process focuses on learning a denoising network that refines the noisy transformation step-by-step, bringing it closer to the optimal transformation for accurate pose estimation. Unlike standard diffusion models used in linear Euclidean spaces, our diffusion model operates on the SE(3) manifold. This requires exploiting the linear Lie algebra $\mathfrak{se}(3)$ associated with SE(3) to constrain the transformation transitions during the diffusion and reverse processes. Additionally, to effectively train our denoising network, we derive a registration-specific variational lower bound as the optimization objective for model learning. Furthermore, we show that our denoising network can be constructed with a surrogate registration model, making our approach applicable to different deep registration networks. Extensive experiments demonstrate that our diffusion registration framework presents outstanding pose estimation performance on the real-world TUD-L, LINEMOD, and Occluded-LINEMOD datasets.

Diffusion Probabilistic Models (DPMs) have recently shown remarkable performance in image generation tasks, which are capable of generating highly realistic images. When adopting DPMs for image restoration tasks, the crucial aspect lies in how to integrate the conditional information to guide the DPMs to generate accurate and natural output, which has been largely overlooked in existing works. In this paper, we present a unified conditional framework based on diffusion models for image restoration. We leverage a lightweight UNet to predict initial guidance and the diffusion model to learn the residual of the guidance. By carefully designing the basic module and integration module for the diffusion model block, we integrate the guidance and other auxiliary conditional information into every block of the diffusion model to achieve spatially-adaptive generation conditioning. To handle high-resolution images, we propose a simple yet effective inter-step patch-splitting strategy to produce arbitrary-resolution images without grid artifacts. We evaluate our conditional framework on three challenging tasks: extreme low-light denoising, deblurring, and JPEG restoration, demonstrating its significant improvements in perceptual quality and the generalization to restoration tasks. The code will be released at https://zhangyi-3.github.io/project/UCDIR/.

In this paper, we present continuous parametric optical flow, a parametric representation of dense and continuous motion over arbitrary time interval. In contrast to existing discrete-time representations (i.e., flow in between consecutive frames), this new representation transforms the frame-to-frame pixel correspondences to dense continuous flow. In particular, we present a temporal-parametric model that employs B-splines to fit point trajectories using a limited number of frames. To further improve the stability and robustness of the trajectories, we also add an encoder with a neural ordinary differential equation (NODE) to represent features associated with specific times. We also contribute a synthetic dataset and introduce two evaluation perspectives to measure the accuracy and robustness of continuous flow estimation. Benefiting from the combination of explicit parametric modeling and implicit feature optimization, our model focuses on motion continuity and outperforms the flow-based and point-tracking approaches for fitting long-term and variable sequences.