AAAI 2025 Papers

Influence maximization is a key topic in data mining, with broad applications in social network analysis and viral marketing. In recent years, researchers have increasingly turned to machine learning techniques to address this problem. By learning the underlying diffusion processes from data, these methods improve the generalizability of solutions while optimizing objectives to identify the optimal seed set for maximizing influence. Nonetheless, two fundamental challenges remain unresolved: (1) While Graph Neural Networks (GNNs) are increasingly employed to learn diffusion models, their traditional architectures often fail to capture the complex dynamics of influence diffusion, (2) Designing optimization objectives is inherently difficult due to the combinatorial explosion associated with solving this problem. To address these challenges, we propose a novel framework, DeepSN. Our framework employs sheaf neural diffusion to learn diverse influence patterns in a data-driven, end-to-end manner, providing enhanced separability in capturing diffusion characteristics. We also propose an optimization technique that accounts for overlapping influence between vertices, significantly reducing the search space and facilitating the identification of the optimal seed set efficiently. Finally, we conduct extensive experiments on both synthetic and real-world datasets to demonstrate the effectiveness of our framework.

Large amounts of missing data are becoming increasingly ubiquitous in modern high-dimensional datasets. Unfortunately, classical completion methods like low-rank, high-rank, or deep matrix completion (LRMC/HRMC/DMC) are often unable to handle real data that does not fall under their respective models. Here we propose a novel completion strategy that generalizes all these models. The main idea is to find a Union of Subspaces (UoS) that can fit a non-linear embedding of the original data, and complete the data according to this latent UoS. This embedding is obtained through a novel pseudo-completion layer in a deep architecture, and the UoS structure is identified in closed-form through an intermediate clustering layer. Our design reduces the exponential memory requirements that are typically induced by uneven patterns of missing data. We give exact details of our architecture, model, loss functions, and training strategy. Our experiments on over 10 real datasets show that our method consistently outperforms the state-of-the-art accuracy by more than a staggering 40%.

A poster from a long input document can be considered as a one-page easy-to-read multimodal (text and images) summary presented on a nice template with good design elements. Automatic transformation of a long document into a poster is a very less studied but challenging task. It involves content summarization of the input document followed by template generation and harmonization. In this work, we propose a novel deep submodular function which can be trained on ground truth summaries to extract multimodal content from the document and explicitly ensures good coverage, diversity and alignment of text and images. Then, we use an LLM based paraphraser and propose to generate a template with various design aspects conditioned on the input content. We show the merits of our approach through extensive automated and human evaluations.

Understanding relations arising out of interactions among entities can be very difficult, and predicting them is even more challenging. This problem has many applications in various fields, such as financial networks and e-commerce. These relations can involve much more complexities than just involving more than two entities. One such scenario is evolving recursive relations between multiple entities, and so far, this is still an open problem. This work addresses the problem of forecasting higher-order interaction events that can be multi-relational and recursive. We pose the problem in the framework of representation learning of temporal hypergraphs that can capture complex relationships involving multiple entities. The proposed model, \textit{Relational Recursive Hyperedge Temporal Point Process} (RRHyperTPP) uses an encoder that learns a dynamic node representation based on the historical interaction patterns and then a hyperedge link prediction-based decoder to model the occurrence of interaction events. These learned representations are then used for downstream tasks involving forecasting the type and time of interactions. The main challenge in learning from hyperedge events is that the number of possible hyperedges grows exponentially with the number of nodes in the network. This will make the computation of negative log-likelihood of the temporal point process expensive, as the calculation of survival function requires a summation over all possible hyperedges. In our work, we develop a noise contrastive estimation method to learn the parameters of our model, and we have experimentally shown that our models perform better than previous state-of-the-art methods for interaction forecasting.

The ability to estimate temporal relationships is critical for both animals and artificial agents. Cognitive science and neuroscience provide remarkable insights into behavioral and neural aspects of temporal credit assignment. In particular, scale invariance of learning dynamics, observed in behavior and supported by neural data, is one of the key principles that governs animal perception: proportional rescaling of temporal relationships does not alter the overall learning efficiency. Here we integrate a computational neuroscience model of scale invariant memory into deep reinforcement learning (RL) agents. We first provide a theoretical analysis and then demonstrate through experiments that such agents can learn robustly across a wide range of temporal scales, unlike agents built with commonly used recurrent memory architectures such as LSTM. This result illustrates that incorporating computational principles from neuroscience and cognitive science into deep neural networks can enhance adaptability to complex temporal dynamics, mirroring some of the core properties of human learning.

Reinforcement learning (RL), particularly its combination with deep neural networks referred to as deep RL (DRL), has shown tremendous promise across a wide range of applications, suggesting its potential for enabling the development of sophisticated robotic behaviors. Robotics problems, however, pose fundamental difficulties for the application of RL, stemming from the complexity and cost of interacting with the physical world. These challenges notwithstanding, recent advances have enabled DRL to succeed at some real-world robotic tasks. However, state-of-the-art DRL solutions’ maturity varies significantly across robotic applications. In this talk, I will review the current progress of DRL in real-world robotic applications based on our recent survey paper (with Tang, Abbatematteo, Hu, Chandra, and Martı́n-Martı́n), with a particular focus on evaluating the real-world successes achieved with DRL in realizing several key robotic competencies, including locomotion, navigation, stationary manipulation, mobile manipulation, human-robot interaction, and multi-robot interaction. The analysis aims to identify the key factors underlying those exciting successes, reveal underexplored areas, and provide an overall characterization of the status of DRL in robotics. I will also highlight several important avenues for future work, emphasizing the need for stable and sample-efficient real-world RL paradigms, holistic approaches for discovering and integrating various competencies to tackle complex long-horizon, open-world tasks, and principled development and evaluation procedures. The talk is designed to offer insights for RL practitioners and roboticists toward harnessing RL’s power to create generally capable real-world robotic systems.

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.

Non-Rigid Structure-from-Motion (NRSfM) is a classic 3D vision problem, where a 2D sequence is taken as input to estimate the corresponding 3D sequence. Recently, the deep neural networks have greatly advanced the task of NRSfM. However, existing deep NRSfM methods still have limitations in handling the inherent sequence property and motion ambiguity associated with the NRSfM problem. In this paper, we revisit deep NRSfM from two perspectives to address the limitations of current deep NRSfM methods : (1) canonicalization and (2) sequence modeling. We propose an easy-to-implement per-sequence canonicalization method as opposed to the previous per-dataset canonicalization approaches. With this in mind, we propose a sequence modeling method that combines temporal information and subspace constraint. As a result, we have achieved a more optimal NRSfM reconstruction pipeline compared to previous efforts. The effectiveness of our method is verified by testing the sequence-to-sequence deep NRSfM pipeline with corresponding regularization modules on several commonly used datasets.

Current deep multi-modal graph clustering methods primarily rely on Graph Neural Network (GNN) to fully exploit attribute features and graph structures, including message propagation and low-dimensional feature embedding. However, these methods lack further exploration of graph structural information, such as the relationship between nodes and shortest paths. Additionally, they may not sufficiently mine complementary information among multi-modal graph data. To address these issues, we propose a novel Deep Multi-modal Graph Clustering via Graph Transformer Network method, called DMGC-GTN. This method thoroughly dissects and utilizes graph structural information, applying graph smoothing to node features and incorporating various forms of embeddings into the transformer architecture. This achieves a unified embedding of graph structure and multi-modal feature attributes, fully exploiting the complementary information within multi-modal graph data. Extensive experiments demonstrate the effectiveness of our algorithm.

Implicit imitation reinforcement learning (IIRL) is a framework that aims to aid a trainee agent’s learning process via observing the state transitions of a mentor, but without access to the latter's action information. Standard IIRL assumes a shared Markov decision process (MDP) between the mentor and trainee, consequently implying an identical action space. This restriction imposes limitations on the applicability of implicit imitation frameworks in real-life scenarios where, possibly due to variations in physical characteristics, the mentor agent may possess distinct own actions, thereby creating a heterogeneous action setting. In this work, we extend the deep implicit imitation Q-networks (DIIQN) method -an online, model-free, deep RL algorithm for implicit imitation- to allow for heterogeneous action sets between mentor and trainee agents. Equipped with our heterogeneous actions DIIQN (HA-DIIQN) method, a trainee agent can harvest the benefits of IIRL even in heterogeneous action settings, achieving accelerated learning and outperforming non-optimal mentor agents.

Hypergraphs provide a flexible framework for modeling high-order (complex) interactions among multiple entities, extending beyond traditional pairwise correlations in graph structures. However, deep hypergraph neural networks (HGNNs) often face the challenge of oversmoothing with increasing depth, similar to issues in graph neural networks (GNNs). While oversmoothing in GNNs has been extensively studied, its implications in relation to hypergraphs are less explored. This paper addresses this gap by first theoretically exploring the reasons behind oversmoothing in deep HGNNs. Our novel insights suggest that a spectral-based hypergraph convolution, equipped with both low-pass and high-pass filters, can potentially mitigate these effects. Motivated by these findings, we introduce FrameHGNN, a framework that utilizes framelet-based hypergraph convolutions integrating tight framelet transforms with both low-pass and high-pass components, as well as the commonly used strategies in designing deep GNN architecture: initial residual and identity mappings. The experiment results on diverse benchmark datasets demonstrate that FrameHGNN outperforms several state-of-the-art models, effectively reducing oversmoothing while improving predictive accuracy. Our contributions not only advance the theoretical understanding of deep hypergraph learning but also provide a practical spectral-based approach for HGNNs, emphasizing the design of multifrequency channels.

With the advancement of computer vision, numerous models have been proposed for screening of fundus diseases. However, the recognition of multiple fundus diseases is often hampered by the simultaneous presence of multiple disease types and the confluence of lesion types in fundus images. This paper addresses these challenges by conceptualizing them as multi-level feature fusion and self-supervised disease-indicative feature learning problems. We decode fundus images at various levels of granularity to delineate scenarios wherein multiple diseases and lesions co-occur. To effectively integrate these features, we introduce a hierarchical vision transformer (HVT) that adeptly captures both inter-level and intra-level dependencies. A novel forward-attention module is proposed to enhance the integration of lower-level semantic information into higher semantic layers, thereby enriching the representation of complex features. Additionally, we introduce a novel self-supervised mask-consistent feature learner (MCFL). Unlike traditional mask-autoencoders that reconstruct original images using encoder-decoder structures, MCFL utilizes a teacher-student framework to reconstruct mask-consistent feature maps. In this setup, exponential moving averaging is employed to derive classification-guided features, serving as labels for reconstruction rather than merely reconstructing the original images. This innovative approach facilitates the extraction of disease-indicative features. Extensive experiments demonstrate that our method significantly outperforms existing state-of-the-art models.

Online hashing has attracted much research attention for large-scale image retrieval in a streaming way. The main challenge lies in keeping balance between high retrieval accuracy and low training time. Existing online hashing methods almost rely on shallow models rather than deep networks due to high training costs, because it is unacceptable to update hash functions on an order of hours. In addition, the multi-label supervision information is not fully utilized to guide the hash learning process and the affinity matrix is always fixed once constructed. In this paper, we propose a novel Deep Graph Online Hashing (DGOH) method, which for the first time introduces inductive graph neural networks (GNNs) to realize deep online hashing with acceptable training costs on an order of seconds. Furthermore, we mine the multi-label information of the images by constructing a label network and learn label-wise weights dynamically to help to update the affinity matrix. In addition, we provide a strategy to obtain examples from the old data to solve the catastrophic forgetting problem. An integrated objective function is designed to train the entire architecture. Extensive experiments on two common benchmarks demonstrate that the proposed method achieves up to 13.3% accuracy gains over state-of-the-art baselines and shows competitive performance on training time.

Generative AI has made remarkable progress in addressing various design challenges. One prominent area where generative AI could bring significant value is in engineering design. In particular, selecting an optimal set of components and their interfaces to create a mechanical system that meets design requirements is one of the most challenging and time-consuming tasks for engineers. This configuration design task is inherently challenging due to its categorical nature, multiple design requirements a solution must satisfy, and the reliance on physics simulations for evaluating potential solutions. These characteristics entail solving a combinatorial optimization problem with multiple constraints involving black-box functions. To address this challenge, we propose a deep generative model to predict the optimal combination of components and interfaces for a given design problem. To demonstrate our approach, we solve a gear train synthesis problem by first creating a synthetic dataset using a domain-specific language, a parts catalogue, and a physics simulator. We then train a Transformer-based model using this dataset, named GearFormer, which can not only generate quality solutions on its own, but also augment traditional search methods such as an evolutionary algorithm and Monte Carlo tree search. We show that GearFormer outperforms such search methods on their own in terms of satisfying the specified design requirements with orders of magnitude faster generation time. Additionally, we showcase the benefit of hybrid methods that leverage both GearFormer and search methods, which further improve the quality of the solutions.

Cross-modal hashing provides an efficient solution for retrieval tasks across various modalities, such as images and text. However, most existing methods are deterministic models, which overlook the reliability associated with the retrieved results. This omission renders them unreliable for determining matches between data pairs based solely on Hamming distance. To bridge the gap, in this paper, we propose a novel method called Deep Evidential Cross-modal Hashing (DECH). This method equips hashing models with the ability to quantify the reliability level of the association between a query sample and each corresponding retrieved sample, bringing a new dimension of reliability to the cross-modal retrieval process. To achieve this, our method addresses two key challenges: i) To leverage evidential theory in guiding the model to learn hash codes, we design a novel evidence acquisition module to collect evidence and place the evidence captured by hash codes on a Beta distribution to derive a binomial opinion. Unlike existing evidential learning approaches that rely on classifiers, our method collects evidence directly through hash codes. ii) To tackle the task-oriented challenge, we first introduce a method to update the derived binomial opinion, allowing it to present the uncertainty caused by conflicting evidence. Following this manner, we present a strategy to precisely evaluate the reliability level of retrieved results, culminating in performance improvement. We validate the efficacy of our DECH through extensive experimentation on four benchmark datasets. The experimental results demonstrate our superior performance compared to 12 state-of-the-art methods.

Proxy-based metric learning has enhanced semantic similarity with class representatives and exhibited noteworthy performance in deep metric learning (DML) tasks. While these methods alleviate computational demands by learning instance-to-class relationships rather than instance-to-instance relationships, they often limit features to be class-specific, thereby degrading generalization performance for unseen class. In this paper, we introduce a novel perspective called Disentangled Deep Metric Learning (DDML), grounded in the framework of information bottleneck, which applies class-agnostic regularization to existing DML methods. Unlike conventional NormSoftmax methods, which primarily emphasize distinct class-specific features, our DDML enables a diverse feature representation by seamlessly transitioning between class-specific features with the aid of class-agnostic features. It smooths decision boundaries, allowing unseen classes to have stable semantic representations in the embedding space. To achieve this, we learn disentangled representations of both class-specific and class-agnostic features in the context of DML. Empirical results demonstrate that our method addresses the limitations of conventional approaches. Our method easily integrates into existing proxy-based algorithms, consistently delivering improved performance.

Neural Radiance Fields (NeRF) has been widely used in computer vision and graphics, achieving impressive results in novel view synthesis and multi-view 3D reconstruction. However, despite its excellent performance under ideal conditions, NeRF struggles in challenging environments such as hazy, foggy, and underwater scenes, primarily due to the difficulty in decoupling objects from the scattering medium. To mitigate this limitation, we proposed a novel approach for NeRF in scenes with scattering media. Specifically, we leverage pseudo-labels during the early stage of training to guide NeRF in decoupling the densities of objects and the scattering medium, guiding the model toward a more appropriate search space. Furthermore, we introduce a Cyclical Progressive Dimensional Optimization Strategy (CPDOS) that focuses on optimizing a single or a few variables during specific periods. Experimental results demonstrate that our method can effectively simulate hazy and underwater scenes, accurately decouple the scattering medium from objects, estimate atmospheric parameters, and outperform existing methods in novel view synthesis and image restoration tasks.

Large Language Models (LLMs) are known to hallucinate facts and make non-factual statements which can undermine trust in their output. The essence of hallucination lies in the absence of metacognition in LLMs, namely the understanding of their own cognitive processes. However, there has been limited research on quantitatively measuring metacognition within LLMs. Drawing inspiration from cognitive psychology theories, we first quantify the metacognitive ability of LLMs as their ability to evaluate the correctness of responses through confidence. Subsequently, we introduce a general framework called DMC designed to decouple metacognitive ability and cognitive ability. This framework tackles the challenge of noisy quantification caused by the coupling of metacognition and cognition in current research, such as calibration-based metrics. Specifically, the DMC framework comprises two key steps. Initially, the framework tasks the LLM with failure prediction, aiming to evaluate the model's performance in predicting failures, a performance jointly determined by both cognitive and metacognitive abilities of the LLM. Following this, the framework disentangles metacognitive ability and cognitive ability based on the failure prediction performance, providing a quantification of the LLM's metacognitive ability independent of cognitive influences. Experiments conducted on eight datasets across five domains reveal that (1) Our proposed DMC framework effectively separates the metacognition and cognition of LLMs; (2) Various confidence elicitation methods impact the quantification of metacognitve ability differently; (3) Stronger metacognitive ability are exhibited by LLMs with better overall performance; (4) Enhancing metacognition holds promise for alleviating hallucination issues.

Gaussian Splatting has emerged as a prominent 3D representation in novel view synthesis, but it still suffers from appearance variations, which are caused by various factors, such as modern camera ISPs, different time of day, weather conditions, and local light changes. These variations can lead to floaters and color distortions in the rendered images/videos. Recent appearance modeling approaches in Gaussian Splatting are either tightly coupled with the rendering process, hindering real-time rendering, or they only account for mild global variations, performing poorly in scenes with local light changes. In this paper, we propose DAVIGS, a method that decouples appearance variations in a plug-and-play and efficient manner. By transforming the rendering results at the image level instead of the Gaussian level, our approach can model appearance variations with minimal optimization time and memory overhead. Furthermore, our method gathers appearance-related information in 3D space to transform the rendered images, thus building 3D consistency across views implicitly. We validate our method on several appearance-variant scenes, and demonstrate that it achieves state-of-the-art rendering quality with minimal training time and memory usage, without compromising rendering speeds. Additionally, it provides performance improvements for different Gaussian Splatting baselines in a plug-and-play manner.