AAAI 2025 论文列表

Out-of-distribution (OOD) detection, determining whether a given sample is part of the in-distribution (ID) or not, has been newly explored by a generative model-based outlier synthesizing approach, especially with diffusion models. Nonetheless, existing diffusion models often produce outliers that are considerably distant from the ID in pixel-space, showing limited efficacy for capturing subtle distinctions between ID and OOD. To address these issues, we propose a novel framework, Semantic Outlier generation via Nuisance Awareness (SONA), which directly utilizes informative pixel-space ID images in diffusion models. Thereby, the generated outliers achieve two crucial properties: (i) they closely resemble the ID mainly in nuisances, while (ii) represent discriminative semantic information. To facilitate the separate effect on semantics and nuisances, we introduce SONA guidance, providing region-specific guidance. Extensive experiments demonstrate the effectiveness of our framework, achieving an impressive AUROC of 87% on near-OOD datasets, which surpasses the performance of baseline methods by a significant margin of approximately 6%.

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.

Diffusion models represent the state-of-the-art in generative modeling. Due to their high training costs, many works leverage pre-trained diffusion models' powerful representations for downstream tasks, such as face super-resolution (FSR), through fine-tuning or prior-based methods. However, relying solely on priors without supervised training makes it challenging to meet the pixel-level accuracy requirements of discrimination task. Although prior-based methods can achieve high fidelity and high-quality results, ensuring consistency remains a significant challenge. In this paper, we propose a masking strategy with strong and weak constraints and iterative refinement for real-world FSR, termed Diffusion Prior Interpolation (DPI). We introduce conditions and constraints on consistency by masking different sampling stages based on the structural characteristics of the face. Furthermore, we propose a condition Corrector (CRT) to establish a reciprocal posterior sampling process. DPI can balance consistency and diversity and can be seamlessly integrated into pre-trained models. In extensive experiments conducted on synthetic and real datasets, along with consistency validation in face recognition, DPI demonstrates superiority over SOTA FSR methods.

Diffusion Models (DMs) offer robust tools for addressing uncertainty and enhancing adaptability in robotics. This work explores their application to trajectory generation, 3D image synthesis, and interpretable scene understanding. For trajectory planning, we propose using colored Gaussian noise to improve robustness and temporal coherence. In 3D image generation, Transfer Entropy enhances information flow between textual and visual modalities for more coherent outputs. Partial Information Decomposition (PID) is leveraged to improve model interpretability and efficiency in scene generation. Rigorous evaluation will assess trajectory quality, robustness, and real-world transferability, aiming to advance autonomous decision-making and scene understanding in robotics.

In high-stakes domains such as healthcare, finance, and law, the need for explainable AI is critical. Traditional methods for generating attribution maps, including white-box approaches relying on gradients and black-box techniques that perturb inputs, face challenges like gradient vanishing, blurred attributions, and computational inefficiencies. To overcome these limitations, we introduce a novel approach that leverages diffusion models within the framework of Information Bottleneck (IB) theory. By utilizing the Gaussian noise from diffusion models, we connect the information bottleneck with the Minimum Mean Squared Error (MMSE) from classical information theory, enabling precise calculation of mutual information. This connection leads to a new loss function that minimizes the Signal-to-Noise Ratio (SNR), facilitating efficient optimization and producing high-resolution, pixel-level attribution maps. Our method achieves greater clarity and accuracy in attributions than existing techniques, requiring significantly fewer pixel values to reach the necessary predictive confidence. This work demonstrates the power of diffusion models in advancing explainable AI, particularly in identifying critical input features with high precision.

We present Diffusion Model Patching (DMP), a simple method to boost the performance of pre-trained diffusion models that have already reached convergence, with a negligible increase in parameters. DMP inserts a small, learnable set of prompts into the model's input space while keeping the original model frozen. The effectiveness of DMP is not merely due to the addition of parameters but stems from its dynamic gating mechanism, which selects and combines a subset of learnable prompts at every step of the generative process (i.e., reverse denoising steps). This strategy, which we term "mixture-of-prompts'', enables the model to draw on the distinct expertise of each prompt, essentially "patching'' the model's functionality at every step with minimal yet specialized parameters. Uniquely, DMP enhances the model by further training on the original dataset already used for training, even in a scenario where significant improvements are typically not expected due to model convergence. Experiments show that DMP significantly enhances the converged FID of DiT-L/2 on FFHQ by 10.38%, achieved with only a 1.43% parameter increase and 50K additional training iterations.

Large-scale generative models, such as text-to-image diffusion models, have garnered widespread attention across diverse domains due to their creative and high-fidelity image generation. Nonetheless, existing large-scale diffusion models are confined to generating images of up to 1K resolution, which is far from meeting the demands of contemporary commercial applications. Directly sampling higher-resolution images often yields results marred by artifacts such as object repetition and distorted shapes. Addressing the aforementioned issues typically necessitates training or fine-tuning models on higher-resolution datasets. However, this poses a formidable challenge due to the difficulty in collecting large-scale high-resolution images and substantial computational resources. While several preceding works have proposed alternatives to bypass the cumbersome training process, they often fail to produce convincing results. In this work, we probe the generative ability of diffusion models at higher resolution beyond their original capability and propose a novel progressive approach that fully utilizes generated low-resolution images to guide the generation of higher-resolution images. Additionally, we integrate an image sharpening operation into our pipeline, further enhancing image quality. Our method obviates the need for additional training or fine-tuning which significantly lowers the burden of computational costs. Extensive experiments and results validate the efficiency and efficacy of our method.

When trained with severely imbalanced data, deep neural networks often struggle to accurately recognize classes with few samples. Previous studies in long-tailed recognition have attempted to rebalance biased learning using known sample distributions, primarily addressing different classification difficulties at the class level. However, these approaches often overlook the instance difficulty variation within each class. In this paper, we propose a difficulty-aware balancing margin (DBM) loss, which considers both class imbalance and instance difficulty. DBM loss comprises two components: a class-wise margin to mitigate learning bias caused by imbalanced class frequencies, and an instance-wise margin assigned to hard positive samples based on their individual difficulty. DBM loss improves class discriminativity by assigning larger margins to more difficult samples. Our method effortlessly combine with existing approaches and consistently improves performance across various long-tailed recognition benchmarks.

Machine learning (ML) models have been shown to leak private information from their training datasets. Differential Privacy (DP), typically implemented through the differential private stochastic gradient descent algorithm (DP-SGD), has become the standard solution to bound leakage from the models. Despite recent improvements, DP-SGD-based approaches for private learning still usually struggle in the high privacy (ε≤1) and low data regimes, and when the private training datasets are imbalanced. To overcome these limitations, we propose Differentially Private Prototype Learning (DPPL) as a new paradigm for private transfer learning. DPPL leverages publicly pre-trained encoders to extract features from private data and generates DP prototypes that represent each private class in the embedding space and can be publicly released for inference. Since our DP prototypes can be obtained from only a few private training data points and without iterative noise addition, they offer high-utility predictions and strong privacy guarantees even under the notion of pure DP. We additionally show that privacy-utility trade-offs can be further improved when leveraging the public data beyond pre-training of the encoder: in particular, we can privately sample our DP prototypes from the publicly available data points used to train the encoder. Our experimental evaluation with four state-of-the-art encoders, four vision datasets, and under different data and imbalancedness regimes demonstrate DPPL's high performance under strong privacy guarantees in challenging private learning setups.

We study convex optimization problems under differential privacy (DP). With heavy-tailed gradients, existing works achieve suboptimal rates. The main obstacle is that existing gradient estimators have suboptimal tail property, resulting in a superfluous factor of d in the union bound. In this paper, we explore algorithms achieving optimal rates of DP optimization with heavy-tailed gradients. Our first method is a simple clipping approach. Under bounded p-th order moments of gradients, with n samples, it achieves minimax optimal population risk with epsilon less than 1/d. We then propose an iterative updating method, which is more complex but achieves this rate for all epsilon smaller than 1. The results significantly improve over existing methods. Such improvement relies on a careful treatment of the tail behavior of gradient estimators. Our results match the minimax lower bound, indicating that the theoretical limit of stochastic convex optimization under DP is achievable.

Domain adaptive object detection (DAOD) aims to generalize an object detector trained on labeled source-domain data to a target domain without annotations, the core principle of which is source-target feature alignment. Typically, existing approaches employ adversarial learning to align the distributions of the source and target domains as a whole, barely considering the varying significance of distinct regions, say instances under different circumstances and foreground vs background areas, during feature alignment. To overcome the shortcoming, we investigate a differential feature alignment strategy. Specifically, a prediction-discrepancy feedback instance alignment module (dubbed PDFA) is designed to adaptively assign higher weights to instances of higher teacher-student detection discrepancy, effectively handling heavier domain-specific information. Additionally, an uncertainty-based foreground-oriented image alignment module (UFOA) is proposed to explicitly guide the model to focus more on regions of interest. Extensive experiments on widely-used DAOD datasets together with ablation studies are conducted to demonstrate the efficacy of our proposed method and reveal its superiority over other SOTA alternatives.

Differentiable environments have heralded new possibilities for learning control policies by offering rich differentiable information that facilitates gradient-based methods. In comparison to prevailing model-free reinforcement learning approaches, model-based reinforcement learning (MBRL) methods exhibit the potential to effectively harness the power of differentiable information for recovering the underlying physical dynamics. However, this presents two primary challenges: effectively utilizing differentiable information to 1) construct models with more accurate dynamic prediction and 2) enhance the stability of policy training. In this paper, we propose a Differentiable Information Enhanced MBRL method, MB-MIX, to address both challenges. Firstly, we adopt a Sobolev model training approach that penalizes incorrect model gradient outputs, enhancing prediction accuracy and yielding more precise models that faithfully capture system dynamics. Secondly, we introduce mixing lengths of truncated learning windows to reduce the variance in policy gradient estimation, resulting in improved stability during policy learning. To validate the effectiveness of our approach in differentiable environments, we provide theoretical analysis and empirical results. Notably, our approach outperforms previous model-based and model-free methods, in multiple challenging tasks involving controllable rigid robots such as humanoid robots' motion control and deformable object manipulation.

Marked temporal point processes (MTPPs) have been shown to be extremely effective in modeling continuous time event sequences (CTESs). In this work, we present adversarial attacks designed specifically for MTPP models. A key criterion for a good adversarial attack is its imperceptibility. For objects such as images or text, this is often achieved by bounding perturbation in some fixed Lp norm-ball. However, similarly minimizing distance norms between two CTESs in the context of MTPPs is challenging due to their sequential nature and varying time-scales and lengths. We address this challenge by first permuting the events and then incorporating the additive noise to the arrival timestamps. However, the worst case optimization of such adversarial attacks is a hard combinatorial problem, requiring exploration across a permutation space that is factorially large in the length of the input sequence. As a result, we propose a novel differentiable scheme - PERMTPP - using which we can perform adversarial attacks by learning to minimize the likelihood, while minimizing the distance between two CTESs. Our experiments on four real-world datasets demonstrate the offensive and defensive capabilities, and lower inference times of PERMTPP.

The field of Autonomous Driving (AD) has witnessed significant progress in recent years. Among the various challenges faced, the safety evaluation of autonomous vehicles (AVs) stands out as a critical concern. Traditional evaluation methods are both costly and inefficient, often requiring extensive driving mileage in order to encounter rare safety-critical scenarios, which are distributed on the long tail of the complex real-world driving landscape. In this paper, we propose a unified approach, Diffusion-Based Safety-Critical Scenario Generation (DiffScene), to generate high-quality safety-critical scenarios which are both realistic and safety-critical for efficient AV evaluation. In particular, we propose a diffusion-based generation framework, leveraging the power of approximating the distribution of low-density spaces for diffusion models. We design several adversarial optimization objectives to guide the diffusion generation under predefined adversarial budgets. These objectives, such as safety-based objective, functionality-based objective, and constraint-based objective, ensure the generation of safety-critical scenarios while adhering to specific constraints. Extensive experimentation has been conducted to validate the efficacy of our approach. Compared with 6 SOTA baselines, DiffScene generates scenarios that are (1) more safety-critical under 3 metrics, (2) more realistic under 5 distance functions, and (3) more transferable to different AV algorithms. In addition, we demonstrate that training AV algorithms with scenarios generated by DiffScene leads to significantly higher performance in terms of the safety-critical metrics compared to baselines. These findings highlight the potential of DiffScene in addressing the challenges of AV safety evaluation, paving the way for safer AV development.

Image retouching aims to enhance the visual quality of photos. Considering the different aesthetic preferences of users, the target of retouching is subjective. However, current retouching methods mostly adopt deterministic models, which not only neglects the style diversity in the expert-retouched results and tends to learn an average style during training, but also lacks sample diversity during inference. In this paper, we propose a diffusion-based method, named DiffRetouch. Thanks to the excellent distribution modeling ability of diffusion, our method can capture the complex fine-retouched distribution covering various visual-pleasing styles in the training data. Moreover, four image attributes are made adjustable to provide a user-friendly editing mechanism. By adjusting these attributes in specified ranges, users are allowed to customize preferred styles within the learned fine-retouched distribution. Additionally, the affine bilateral grid and contrastive learning scheme are introduced to handle the problem of texture distortion and control insensitivity respectively. Extensive experiments have demonstrated the superior performance of our method on visually appealing and sample diversity.

In social networks, people influence each other through social links, which can be represented as propagation among nodes in graphs. Influence minimization (IMIN) is the problem of manipulating the structures of an input graph (e.g., removing edges) to reduce the propagation among nodes. IMIN can represent time-critical real-world applications, such as rumor blocking, but IMIN is theoretically difficult and computationally expensive. Moreover, the discrete nature of IMIN hinders the usage of powerful machine learning techniques, which requires differentiable computation. In this work, we propose DiffIM, a novel method for IMIN with two differentiable schemes for acceleration: (1) surrogate modeling for efficient influence estimation, which avoids time-consuming simulations (e.g., Monte Carlo), and (2) the continuous relaxation of decisions, which avoids the evaluation of individual discrete decisions (e.g., removing an edge). We further propose a third accelerating scheme, gradient-driven selection, that chooses edges instantly based on gradients without optimization (spec., gradient descent iterations) on each test instance. Through extensive experiments on real-world graphs, we show that each proposed scheme significantly improves speed with little (or even no) IMIN performance degradation. Our method is Pareto-optimal (i.e., no baseline is faster and more effective than it) and typically several orders of magnitude (spec., up to 15,160X) faster than the most effective baseline, while being more effective.

Generating high-quality whole-body human object interaction motion sequences is becoming increasingly important in various fields such as animation, VR/AR, and robotics. The main challenge of this task lies in determining the level of involvement of each hand given the complex shapes of objects in different sizes and their different motion trajectories, while ensuring strong grasping realism and guaranteeing the coordination of movement in all body parts. Contrasting with existing work, which either generates human interaction motion sequences without detailed hand grasping poses or only models a static grasping pose, we propose a simple yet effective framework that jointly models the relationship between the body, hands, and the given object motion sequences within a single diffusion model. To guide our network in perceiving the object's spatial position and learning more natural grasping poses, we introduce novel contact-aware losses and incorporate a data-driven, carefully designed guidance. Experimental results demonstrate that our approach outperforms the state-of-the-art method and generates plausible results.

Fine-tuning text-to-image diffusion models to maximize rewards has proven effective for enhancing model performance. However, reward fine-tuning methods often suffer from slow convergence due to online sample generation. Therefore, obtaining diverse samples with strong reward signals is crucial for improving sample efficiency and overall performance. In this work, we introduce DiffExp, a simple yet effective exploration strategy for reward fine-tuning of text-to-image models. Our approach employs two key strategies: (a) dynamically adjusting the scale of classifier-free guidance to enhance sample diversity, and (b) randomly weighting phrases of the text prompt to exploit high-quality reward signals. We demonstrate that these strategies significantly enhance exploration during online sample generation, improving the sample efficiency of recent reward fine-tuning methods, such as DDPO and AlignProp.

Dense video captioning (DVC) aims to describe multiple events within a video, and its performance is greatly affected by the accuracy of video event detection. Video event detection involves predicting the proposal boundaries (start and end times) and the classification score of each event in a video. Recently, a few methods have applied diffusion models originally designed for image object detection to detect events in DVC. These methods add noise to the ground-truth event proposal boundaries, and subsequently learn the denoising process. However, these methods often overlook the fundamental differences between videos and images. We observe that, whereas in images the important information for object classification is normally around the boundaries of the ground-truth boxes, in videos the key information for event classification is typically centered in the middle of ground-truth event proposals. As a result, the classification module in these existing diffusion models becomes insensitive to boundary changes introduced by the added noise, leading to sub-optimal performance. This paper introduces DiffDVC, an innovative diffusion model for DVC. The core of DiffDVC is a boundary-sensitive detector. The detector increases the sensitivity of the classification module to boundary changes by focusing on frames within a specific range around the start and end times of noisy event proposals. Additionally, this range is dynamically adjusted to suit different event proposals. Comprehensive experiments on ActivityNet-1.3, ActivityNet Captions, and YouCook2 datasets show DiffDVC achieving superior performance.