AAAI 2025 论文列表

Modeling 3D open-vocabulary language fields is challenging yet highly anticipated. Despite great progress, existing approaches heavily rely on a large number of training views to construct language-embedded 3D scenes, which is unfortunately impractical in real-world scenarios. This paper introduces SOVGaussian, the first method for few-shot novel view open-vocabulary language querying. We introduce a depth-constrained neural language field to mitigate the geometry degradation caused by overfitting training views. Rather than straightforwardly using dense depth maps for loosely accurate supervision, Language-Aware Depth Distillation (LAD) based on open-vocabulary object masks is proposed, ensuring intra-object geometric accuracy within the language field. To further refine the language-geometry consistency of the language field, we propose a novel Language-Guided Outlier Pruning (LOP) strategy, which identifies floating 3D Gaussian primitives overfitting training views based on their language-grouped densities. Our comprehensive experiments demonstrate that SOVGaussian is able to reconstruct a superior scene representation from few-shot images, outperforming existing state-of-the-art methods and achieving significantly better performance on novel view language querying and synthesis.

Mixed Integer Linear Program (MILP) solvers are mostly built upon a Branch-and-Bound (B&B) algorithm, where the efficiency of traditional solvers heavily depends on hand-crafted heuristics for branching. The past few years have witnessed the increasing popularity of data-driven approaches to automatically learn these heuristics. However, the success of these methods is highly dependent on the availability of high-quality demonstrations, which requires either the development of near-optimal heuristics or a time-consuming sampling process. This paper averts this challenge by proposing Suboptimal-Demonstration-Guided Reinforcement Learning (SORREL) for learning to branch. SORREL selectively learns from suboptimal demonstrations based on value estimation. It utilizes suboptimal demonstrations through both offline reinforcement learning on the demonstrations generated by suboptimal heuristics and self-imitation learning on past good experiences sampled by itself. Our experiments demonstrate its advanced performance in both branching quality and training efficiency over previous methods for various MILPs.

Graph contrastive learning has emerged as a powerful technique for learning graph representations that are robust and discriminative. However, traditional approaches often neglect the critical role of subgraph structures, particularly the intra-subgraph characteristics and inter-subgraph relationships, which are crucial for generating informative and diverse contrastive pairs. These subgraph features are crucial as they vary significantly across different graph types, such as social networks where they represent communities, and biochemical networks where they symbolize molecular interactions. To address this issue, our work proposes a novel subgraph-oriented learnable augmentation method for graph contrastive learning, termed SOLA-GCL, that centers around subgraphs, taking full advantage of the subgraph information for data augmentation. Specifically, SOLA-GCL initially partitions a graph into multiple densely connected subgraphs based on their intrinsic properties. To preserve and enhance the unique characteristics inherent to subgraphs, a graph view generator optimizes augmentation strategies for each subgraph, thereby generating tailored views for graph contrastive learning. This generator uses a combination of intra-subgraph and inter-subgraph augmentation strategies, including node dropping, feature masking, intra-edge perturbation, inter-edge perturbation, and subgraph swapping. Extensive experiments have been conducted on various graph learning applications, ranging from social networks to molecules, under semi-supervised learning, unsupervised learning, and transfer learning settings to demonstrate the superiority of our proposed approach.

Predicting the next activity in an ongoing process is one of the most common tasks in the business process management (BPM) domain. It allows businesses to optimize resource allocation, enhance operational efficiency, and aid both in risk mitigation and strategic decision-making. Existing state-of-the-art AI models for BPM do not fully capitalize on available semantic information within process event logs. As current advanced AI-BPM systems provide semantically richer textual data, the need for new adequate models grows. To address this gap, we develop SNAP—a novel system that utilizes LLMs by constructing narratives and semantic contextual stories for historical event logs, which are then used to generate precise and actionable predictions for the ongoing process. SNAP was evaluated on six benchmark datasets, where it demonstrated significant performance improvements over eleven SOTA models, particularly on datasets with high levels of semantic content. This work showcases the potential of integrating LLMs in BPM and outlines a clear path toward future deployment, emphasizing the relevance and innovation of our approach within the broader AI application landscape.

Continuous control tasks often involve high-dimensional, dynamic, and non-linear environments. State-of-the-art performance in these tasks is achieved through complex closed-box policies that are effective, but suffer from an inherent opacity. Interpretable policies, while generally underperforming compared to their closed-box counterparts, advantageously facilitate transparent decision-making within automated systems. Hence, their usage is often essential for diagnosing and mitigating errors, supporting ethical and legal accountability, and fostering trust among stakeholders. In this paper, we propose SMoSE, a novel method to train sparsely activated interpretable controllers, based on a top-1 Mixture-of-Experts architecture. SMoSE combines a set of interpretable decision-makers, trained to be experts in different basic skills, and an interpretable router that assigns tasks among the experts. The training is carried out via state-of-the-art Reinforcement Learning algorithms, exploiting load-balancing techniques to ensure fair expert usage. We then distill decision trees from the weights of the router, significantly improving the ease of interpretation. We evaluate SMoSE on six benchmark environments from MuJoCo: our method outperforms recent interpretable baselines and narrows the gap with non-interpretable state-of-the-art algorithms.

Transformer-based methods have achieved remarkable performance in event-based object detection, owing to the global modeling ability. However, they neglect the influence of non-event and noisy regions and process them uniformly, leading to high computational overhead. To mitigate computation cost, some researchers propose window attention based sparsification strategies to discard unimportant regions, which sacrifices the global modeling ability and results in suboptimal performance. To achieve better trade-off between accuracy and efficiency, we propose Sparse Mamba (SMamba), which performs adaptive sparsification to reduce computational effort while maintaining global modeling capability. Specifically, a Spatio-Temporal Continuity Assessment module is proposed to measure the information content of tokens and discard uninformative ones by leveraging the spatiotemporal distribution differences between activity and noise events. Based on the assessment results, an Information-Prioritized Local Scan strategy is designed to shorten the scan distance between high-information tokens, facilitating interactions among them in the spatial dimension. Furthermore, to extend the global interaction from 2D space to 3D representations, a Global Channel Interaction module is proposed to aggregate channel information from a global spatial perspective. Results on three datasets (Gen1, 1Mpx, and eTram) demonstrate that our model outperforms other methods in both performance and efficiency.

This paper introduces a lightweight Semantic-guided Mutually Reinforcing network (SMR-Net) for the tasks of cross-modal image fusion and salient object detection (SOD). The core concept of SMR-Net is to leverage semantics for directing the mutual reinforcing between image fusion and SOD. Specifically, a Progressive Cross-modal Interaction (PCI) image fusion subnetwork is designed to exploit local interactions via convolution operations and extend to global interactions utilizing spatial and channel attention mechanisms. Subsequently, a cross-modal Bit-Plane Slicing-based SOD subnetwork (BPS) is developed by incorporating the fused image as a third modality. This component employs bit-plane slicing and the deformable convolution technique to effectively extract irregular semantic information embedded in fusion features. The refined semantic information then guides the feature extraction process of the source modalities in a reweighted fashion. By cascading these two subnetworks, BPS leverages final semantic results to direct PCI towards focusing more on semantic information. Ultimately, through this semantic-guided mutual enhancement process, SMR-Net excels in both producing high-quality fused images and achieving effective salient object detection. Our extensive experiments on image fusion and SOD tasks convincingly demonstrate the superiority of our network over existing state-of-the-art alternatives without introducing noticeable computational costs. Compared to nearest competitors, our method demonstrates a stronger generalization ability with 26% fewer parameters.

We propose SMMF (Square-Matricized Momentum Factorization), a memory-efficient optimizer that reduces the memory requirement of the widely used adaptive learning rate optimizers, such as Adam, by up to 96%. SMMF enables flexible and efficient factorization of an arbitrary rank (shape) of the first and second momentum tensors during optimization, based on the proposed square-matricization and one-time single matrix factorization. From this, it becomes effectively applicable to any rank (shape) of momentum tensors, i.e., bias, matrix, and any rank-d tensors, prevalent in various deep model architectures, such as CNNs (high rank) and Transformers (low rank), in contrast to existing memory-efficient optimizers that applies only to a particular (rank-2) momentum tensor, e.g., linear layers. We conduct a regret bound analysis of SMMF, which shows that it converges similarly to non-memory-efficient adaptive learning rate optimizers, such as AdamNC, providing a theoretical basis for its competitive optimization capability. In our experiment, SMMF takes up to 96% less memory compared to state-of-the-art memoryefficient optimizers, e.g., Adafactor, CAME, and SM3, while achieving comparable model performance on various CNN and Transformer tasks.

Despite the extent of recent advances in Machine Learning (ML) and Neural Networks, providing formal guarantees on the behavior of these systems is still an open problem, and a crucial requirement for their adoption in regulated or safety-critical scenarios. We consider the task of training differentiable ML models guaranteed to satisfy designer-chosen properties, stated as input-output implications. This is very challenging, due to the computational complexity of rigorously verifying and enforcing compliance in deep neural models. We provide an innovative approach based on: 1) a general, simple architecture enabling efficient verification with a conservative semantic; 2) a rigorous training algorithm based on the Projected Gradient Method; 3) a formulation of the problem of searching for strong counterexamples. The proposed framework, being only marginally affected by model complexity, scales well to practical applications, and produces models that provide full property satisfaction guarantees. We evaluate our approach on properties defined by linear inequalities in regression, and on mutually exclusive classes in multi-label classification. Our approach is competitive with a baseline that includes property enforcement in preprocessing (on training data) and postprocessing (on model predictions). Finally, our contributions establish a framework that opens up multiple research directions and potential improvements.

Most existing semi-supervised community detection algorithms leverage known communities to learn community structures, subsequently identifying communities that align with these learned community structures. However, differences in community structures may render the community structures learned by these methods inappropriate for the community containing the given node of interest. As a result, the identified community may exclude the given node or be of poor quality. Inspired by the success of reinforcement learning, we propose a Semi-supervised Local community detection method based on Reinforcement Learning, named SLRL, which only explores parts of the network surrounding the given node. It first extracts the local structure around a given node with an extractor, followed by selecting communities that are similar to this local structure to distill useful communities. These selected communities are employed to train the expander, which expands the community containing a given node. Experimental results demonstrate that SLRL outperforms state-of-the-art algorithms on five real-world datasets.

Multi-view tensor clustering (MVTC) has gained much attention for its effectiveness in capturing global high-order correlations across views. However, current MVTC methods suffer from two limitations: 1) adopting a two-stage process to learn the latent features for clustering, and 2) either ignoring local similarities within views or treating local similarities and global high-order correlations equally. In this paper, we propose a smooth low-rank MVTC (SLR-MVTC) method, which aims to extract latent features that are smooth within each view and low-rank across views, enhancing clustering performance. Specifically, we first learn latent features from each view using orthogonal projection and then construct the latent feature tensor by concatenation and rotation. Then, we introduce a new smooth tensor nuclear norm to depict the low-rank components of the low-frequency parts in the feature tensor. Benefiting from the fast Fourier transform along the sample dimension, the obtained low-frequency components effectively capture local smoothness within views, while their low-rank parts further explore global correlations across views. Experimental results on six multi-view datasets demonstrate that SLR-MVTC outperforms state-of-the-art algorithms in terms of clustering performance and CPU time.

Face anti-spoofing (FAS) plays a pivotal role in ensuring the security and reliability of face recognition systems. With advancements in vision-language pretrained (VLP) models, recent two-class FAS techniques have leveraged the advantages of using VLP guidance, while this potential remains unexplored in one-class FAS methods. The one-class FAS focuses on learning intrinsic liveness features solely from live training images to differentiate between live and spoof faces. However, the lack of spoof training data can lead one-class FAS models to inadvertently incorporate domain information irrelevant to the live/spoof distinction (\eg, facial content), causing performance degradation when tested with a new application domain. To address this issue, we propose a novel framework called Spoof-aware one-class face anti-spoofing with Language Image Pretraining (SLIP). Given that live faces should ideally not be obscured by any spoof-attack-related objects (\eg, paper, or masks) and are assumed to yield zero spoof cue maps, we first propose an effective language-guided spoof cue map estimation to enhance one-class FAS models by simulating whether the underlying faces are covered by attack-related objects and generating corresponding nonzero spoof cue maps. Next, we introduce a novel prompt-driven liveness feature disentanglement to alleviate live/spoof-irrelative domain variations by disentangling live/spoof-relevant and domain-dependent information. Finally, we design an effective augmentation strategy by fusing latent features from live images and spoof prompts to generate spoof-like image features and thus diversify latent spoof features to facilitate the learning of one-class FAS. Our extensive experiments and ablation studies support that SLIP consistently outperforms previous one-class FAS methods.

Ordinal regression classifies an object to a class out of a given set of possible classes, where labels possess a natural order. It is relevant to a wide array of domains including risk assessment, sentiment analysis, image ranking, and recommender systems. Like common classification, the primary goal of ordinal regression is accuracy. Yet, in this context, the severity of prediction errors varies, e.g., in risk assessment, Critical Risk is more urgent than High risk and significantly more urgent than No risk. This leads to a modified objective of ensuring that the model's output is as close as possible to the correct class, considering the order of labels. Therefore, ordinal regression models should use ordinality-aware loss for training. In this work, we focus on two properties of ordinality-aware losses, namely monotonicity and balance sensitivity. We show that existing ordinal loss functions lack these properties and introduce SLACE (Soft Labels Accumulating Cross Entropy), a novel loss function that provably satisfies said properties. We demonstrate empirically that SLACE outperforms the state-of-the-art ordinal loss functions on most tabular ordinal regression benchmarks.

The introduction of vision-language models like CLIP has enabled the development of foundational video models capable of generalizing to unseen videos and human actions. However, these models are typically trained on web videos, which often fail to capture the challenges present in Activities of Daily Living (ADL) videos. Existing works address ADL-specific challenges, such as similar appearances, subtle motion patterns, and multiple viewpoints, by combining 3D skeletons and RGB videos. However, these approaches are not integrated with language, limiting their ability to generalize to unseen action classes. In this paper, we introduce SKI models, which integrate 3D skeletons into the vision-language embedding space. SKI models leverage a skeleton-language model, SkeletonCLIP, to infuse skeleton information into Vision Language Models (VLMs) and Large Vision Language Models (LVLMs) through collaborative training. Notably, SKI models do not require skeleton data during inference, enhancing their robustness for real-world applications. The effectiveness of SKI models is validated on three popular ADL datasets for zero-shot action recognition and video caption generation tasks.

An efficient and precise diagnosis of retinal diseases is a fundamental goal for auxiliary diagnostic systems in ophthalmology. Inspired by the importance of scattered subtle lesions in manual retinal disease diagnosis, recent research has achieved state-of-the-art performance by mining information related to subtle lesions, including their texture and shape. However, the spatial distribution patterns of subtle lesion areas, which are also crucial in manual diagnosis, have been overlooked in existing research. Neglecting these spatial distribution patterns (e.g., the ring distribution of microaneurysms in diabetic macular edema) may negatively impact the diagnostic process. In this paper, we introduce the Saliency-Image-Graph (SIGraph) network to capture the spatial distribution patterns of lesion areas. We first employ saliency-based perception to identify latent lesion pixels. Subsequently, we propose a novel image-graph block to efficiently capture the global distribution of abundant lesion pixels with minimal information loss. By leveraging additional distribution patterns, SIGraph achieves state-of-the-art performance with at least a 1.5% performance gain across three datasets. Furthermore, ablation studies demonstrate that our image-graph block can be integrated into other visual backbones and effectively boost performance.

Sequential Recommender Systems (SRS) has stood out as a highly promising technique in numerous domains due to its impressive capability of capturing complex user preferences. Current SRS have employed transformer-based models to give the next-item prediction. Nevertheless, its quadratic computational complexity has often resulted in notable inefficiencies, posing a significant obstacle to real-time recommendation processes. Recently, Mamba has demonstrated its exceptional effectiveness in time series prediction, delivering substantial improvements in both efficiency and effectiveness. However, directly applying Mamba to SRS poses certain challenges. Its unidirectional structure may impede the ability to capture contextual information in user-item interactions, while its instability in state estimation may hinder the ability to capture short-term patterns in interaction sequences. To address these issues, we propose a novel framework called Selective Gated Mamba for Sequential Recommendation (SIGMA). By introducing the Partially Flipped Mamba (PF-Mamba), we construct a special bi-directional structure to address the context modeling challenge. Then, to consolidate PF-Mamba's performance, we employed an input-dependent Dense Selective Gate (DS Gate) to allocate the weights of the two directions and further filter the sequential information. Moreover, for short sequence modeling, we devise a Feature Extract GRU (FE-GRU) to capture the short-term dependencies. Experimental results demonstrate that SIGMA significantly outperforms existing baselines across five real-world datasets. Our implementation code is available in Supplementary Material to ease reproducibility.

Smartphone cameras are ubiquitous in daily life, yet their performance can be severely impacted by dirty lenses, leading to degraded image quality. This issue is often overlooked in image restoration research, which assumes ideal or controlled lens conditions. To address this gap, we introduced SIDL (Smartphone Images with Dirty Lenses), a novel dataset designed to restore images captured through contaminated smartphone lenses. SIDL contains diverse real-world images taken under various lighting conditions and environments. These images feature a wide range of lens contaminants, including water drops, fingerprints, and dust. Each contaminated image is paired with a clean reference image, enabling supervised learning approaches for restoration tasks. To evaluate the challenge posed by SIDL, various state-of-the-art restoration models were trained and compared on this dataset. Their performances achieved some level of restoration but did not adequately address the diverse and realistic nature of the lens contaminants in SIDL. This challenge highlights the need for more robust and adaptable image restoration techniques for restoring images with dirty lenses.

Drought has become a critical global threat with significant societal impact. Existing drought monitoring solutions primarily focus on assessing drought severity using quantitative measurements, overlooking the diverse societal impact of drought from human-centric perspectives. Motivated by the collective intelligence on social media and the computational power of AI, this paper studies a novel problem of socially informed AI-driven drought estimation that aims to leverage social and news media information to jointly estimate drought severity and its societal impact. Two technical challenges exist: 1) How to model the implicit temporal dynamics of drought societal impact. 2) How to capture the social-physical interdependence between the physical drought condition and its societal impact. To address these challenges, we develop SIDE, a socially informed AI-driven drought estimation framework that explicitly quantifies the societal impact of drought and effectively models the social-physical interdependency for joint severity-impact estimation. Experiments on real-world datasets from California and Texas demonstrate SIDE's superior performance compared to state-of-the-art baselines in accurately estimating drought severity and its societal impact. SIDE offers valuable insights for developing human-centric drought mitigation strategies to foster sustainable and resilient communities.

Although semi-supervised learning has made significant advances in the field of medical image segmentation, fully annotating a volumetric sample slice by slice remains a costly and time-consuming task. Even worse, most of the existing approaches pay much attention to image-level information and ignore semantic features, resulting in the inability to perceive weak boundaries. To address these issues, we propose a novel Semantic-Guided Triplet Co-training (SGTC) framework, which achieves high-end medical image segmentation by only annotating three orthogonal slices of a few volumetric samples, significantly alleviating the burden of radiologists. Our method consist of two main components. Specifically, to enable semantic-aware, fine-granular segmentation and enhance the quality of pseudo-labels, a novel semantic-guided auxiliary learning mechanism is proposed based on the pretrained CLIP. In addition, focusing on a more challenging but clinically realistic scenario, a new triple-view disparity training strategy is proposed, which uses sparse annotations (i.e., only three labeled slices of a few volumes) to perform co-training between three sub-networks, significantly improving the robustness. Extensive experiments on three public medical datasets demonstrate that our method outperforms most state-of-the-art semi-supervised counterparts under sparse annotation settings.