AAAI 2025 Papers

Present bias, the tendency to overvalue immediate rewards while undervaluing future ones, is a well-known barrier to achieving long-term goals. As artificial intelligence and behavioral economics increasingly focus on this phenomenon, the need for robust mathematical models to predict behavior and guide effective interventions has become crucial. However, existing models are constrained by their reliance on the discreteness of time and limited discount functions. This study introduces a novel continuous-time mathematical model for agents influenced by present bias. Using the variational principle, we model human behavior, where individuals repeatedly act according to a sequence of states that minimize their perceived cost. Our model not only retains analytical tractability but also accommodates various discount functions. Using this model, we consider intervention optimization problems under exponential and hyperbolic discounting and theoretically derive optimal intervention strategies, offering new insights into managing present-biased behavior.

We introduce a novel language for reasoning about agents' cognitive attitudes of both epistemic and motivational type. We interpret it by means of a computationally grounded semantics using belief bases. Our language includes five types of modal operators for implicit belief, complete attraction, complete repulsion, realistic attraction and realistic repulsion. We give an axiomatization and show that our operators are not mutually expressible and that they can be combined to represent a large variety of psychological concepts including ambivalence, indifference, being motivated, being demotivated and preference. We present a dynamic extension of the language that supports reasoning about the effects of belief change operations. Finally, we provide a succinct formulation of model checking for our languages and a PSPACE model checking algorithm relying on a reduction into TQBF. We present some experimental results for the implemented algorithm on computation time in a concrete example.

Multimodal Large Language Models (MLLMs) have showcased impressive skills in tasks related to visual understanding and reasoning. Yet, their widespread application faces obstacles due to the high computational demands during both the training and inference phases, restricting their use to a limited audience within the research and user communities. In this paper, we investigate the design aspects of Multimodal Small Language Models (MSLMs) and propose an efficient multimodal assistant named Mipha, which is designed to create synergy among various aspects: visual representation, language models, and optimization strategies. We show that without increasing the volume of training data, our Mipha-3B outperforms the state-of-the-art large MLLMs, especially LLaVA-1.5-13B, on multiple benchmarks. Through detailed discussion, we provide insights and guidelines for developing strong MSLMs that rival the capabilities of MLLMs.

Event reasoning is a fundamental ability that underlies many applications. It requires event schema knowledge to perform global reasoning and needs to deal with the diversity of the inter-event relations and the reasoning paradigms. The extent to which LLMs excel in event reasoning across various relations and reasoning paradigms has not been thoroughly investigated. Additionally, it is still unclear whether LLMs utilize event knowledge in the same way humans do. To mitigate this disparity, we comprehensively evaluate the abilities of event reasoning of LLMs on different relations, paradigms, and levels of abstraction. We introduce a novel benchmark EV2 for EValuation of EVent reasoning. EV2 consists of two levels of evaluation on schema and instance and is comprehensive in relations and reasoning paradigms. We conduct extensive experiments on EV2. We find that 1) LLMs have abilities to accomplish event reasoning but their performances are far from satisfactory. 2) There are imbalances of event reasoning abilities on different relations and paradigms. 3) LLMs have event schema knowledge, however, they're not aligned with humans on how to utilize the knowledge. Based on these findings, we guide the LLMs in utilizing the event schema knowledge as memory leading to improvements in event reasoning.

Optimization Modulo Nonlinear Real Arithmetic, abbreviated as OMT(NRA), generally focuses on optimizing a given objective subject to quantifier-free Boolean combinations of primitive constraints, including Boolean variables, polynomial equations, and inequalities. It is widely applicable in areas like program verification, analysis, planning, and so on. The existing solver, OptiMathSAT, officially supporting OMT(NRA), employs an incomplete algorithm. We present a sound and complete algorithm, Optimization Cylindrical Algebraic Covering (OCAC), integrated within the Conflict-Driven Clause Learning (CDCL) framework, specifically tailored for OMT(NRA) problems. We establish the correctness and termination of CDCL(OCAC) and explore alternative approaches using cylindrical algebraic decomposition (CAD) and first-order formulations. Our work includes the development of the first complete OMT solver for NRA, demonstrating significant performance improvements. In benchmarks generated from SMT-LIB instances, our algorithm finds the optimum value in about 150% more instances compared to the current leading solver, OptiMathSAT.

Large language models (LLMs) have made significant advancements in math problem solving, but their large size and high latency render them impractical for real-world applications in intelligent mathematics solvers. Recently, task-agnostic compact models have been developed to replace LLMs in general natural language processing tasks. However, these models often struggle to acquire sufficient math-related knowledge from LLMs, leading to unsatisfactory performance in solving math word problems (MWPs). To develop a specialized compact model for representing MWPs, we develop the knowledge distillation (KD) technique to extract mathematical semantics knowledge from the large pre-trained model BERT. Effective knowledge types and distillation strategies are explored through extensive experiments. Our KD algorithm employs multi-knowledge distillation to extract fundamental knowledge from hidden states in the middle to lower layers, while also incorporating knowledge of mathematical relations and symbol constraints from higher-layer outputs and math decoder outputs, by leveraging bottleneck networks. Pre-training tasks on MWP datasets, such as masked language modeling and part-of-speech tagging, are also utilized to enhance the generalization of the compact model for MWP understanding. Additionally, a simple parameter mixing strategy is employed to prevent catastrophic forgetting of acquired knowledge. Our findings indicate that our approach can reduce the size of a BERT model by 10% while retaining approximately 95% of its performance on MWP datasets, outperforming the mainstream BERT-based task-agnostic compact models. The efficacy of each component has been validated through ablation studies.

Functional Magnetic Resonance Imaging (fMRI) data is a widely used kind of four-dimensional biomedical data, which requires effective compression. However, fMRI compressing poses unique challenges due to its intricate temporal dynamics, low signal-to-noise ratio, and complicated underlying redundancies. This paper reports a novel compression paradigm specifically tailored for fMRI data based on Implicit Neural Representation (INR). The proposed approach focuses on removing the various redundancies among the time series by employing several methods, including (i) conducting spatial correlation modeling for intra-region dynamics, (ii) decomposing reusable neuronal activation patterns, and (iii) using proper initialization together with nonlinear fusion to describe the inter-region similarity. This scheme appropriately incorporates the unique features of fMRI data, and experimental results on publicly available datasets demonstrate the effectiveness of the proposed method, surpassing state-of-the-art algorithms in both conventional image quality evaluation metrics and fMRI downstream tasks. This work in this paper paves the way for sharing massive fMRI data at low bandwidth and high fidelity.

Camera-based Bird's Eye View (BEV) perception models receive increasing attention for their crucial role in autonomous driving, a domain where concerns about the robustness and reliability of deep learning have been raised. While only a few works have investigated the effects of randomly generated semantic perturbations, aka natural corruptions, on the multi-view BEV detection task, we develop a black-box robustness evaluation framework that adversarially optimises three common semantic perturbations: geometric transformation, colour shifting, and motion blur, to deceive BEV models, serving as the first approach in this emerging field. To address the challenge posed by optimising the semantic perturbation, we design a smoothed, distance-based surrogate function to replace the mAP metric and introduce SimpleDIRECT, a deterministic optimisation algorithm that utilises observed slopes to guide the optimisation process. By comparing with randomised perturbation and two optimisation baselines, we demonstrate the effectiveness of the proposed framework. Additionally, we provide a benchmark on the semantic robustness of ten recent BEV models. The results reveal that PolarFormer, which emphasises geometric information from multi-view images, exhibits the highest robustness, whereas BEVDet is fully compromised, with its precision reduced to zero.

Efficiently computing accurate representations of high-dimensional data is essential for data analysis and unsupervised learning. Dendrograms, also known as ultrametrics, are widely used representations that preserve hierarchical relationships within the data. However, popular methods for computing them, such as *linkage* algorithms, suffer from quadratic time and space complexity, making them impractical for large datasets. The "best ultrametric embedding" (a.k.a. "best ultrametric fit") problem, which aims to find the ultrametric that best preserves the distances between points in the original data, is known to require at least quadratic time for an exact solution. Recent work has focused on improving scalability by approximating optimal solutions in subquadratic time, resulting in a (sqrt(2) + epsilon)-approximation (Cohen-Addad, de Joannis de Verclos and Lagarde, 2021). In this paper, we present the first subquadratic algorithm that achieves arbitrarily precise approximations of the optimal ultrametric embedding. Specifically, we provide an algorithm that, for any c >1, outputs a c-approximation of the best ultrametric in time O(n^(1 + 1/c)). In particular, for any fixed epsilon > 0, the algorithm computes a (1+ epsilon)-approximation in time O(n^(2 - epsilon + o(epsilon ^2))). Experimental results show that our algorithm improves upon previous methods in terms of approximation quality while maintaining comparable running times.

Protein structure prediction is pivotal for understanding the structure-function relationship of proteins, advancing biological research, and facilitating pharmaceutical development and experimental design. While deep learning methods and the expanded availability of experimental 3D protein structures have accelerated structure prediction, the dynamic nature of protein structures has received limited attention. This study introduces an innovative 4D diffusion model incorporating molecular dynamics (MD) simulation data to learn dynamic protein structures. Our approach is distinguished by the following components: (1) a unified diffusion model capable of generating dynamic protein structures, including both the backbone and side chains, utilizing atomic grouping and side-chain dihedral angle predictions; (2) a reference network that enhances structural consistency by integrating the latent embeddings of the initial 3D protein structures; and (3) a motion alignment module aimed at improving temporal structural coherence across multiple time steps. To our knowledge, this is the first diffusion-based model aimed at predicting protein trajectories across multiple time steps simultaneously. Validation on benchmark datasets demonstrates that our model exhibits high accuracy in predicting dynamic 3D structures of proteins containing up to 256 amino acids over 32 time steps, effectively capturing both local flexibility in stable states and significant conformational changes.

The combination of self-supervised learning and adversarial training (AT) can significantly improve the adversarial robustness of self-supervised models. However, the robustness of self-supervised adversarial training (self-AT) still lags behind that of state-of-the-art (SOTA) supervised AT (sup-AT), even though the performance of current self-supervised learning models has already matched or even surpassed that of SOTA supervised learning models. This issue raises concerns about the secure application of self-supervised learning models. The inclusion of adversarial training turns self-AT into a challenging joint optimization problem, and recent studies have shown that the data augmentation methods necessary for constructing positive pairs in self-supervised learning negatively impact the robustness improvement in self-AT. Inspired by this, we propose 3SAT, a simple self-supervised adversarial training framework. 3SAT conducts adversarial training on original, unaugmented samples, reducing the difficulty of optimizing the adversarial training subproblem and fundamentally eliminating the negative impact of data augmentation on robustness improvement. Additionally, 3SAT introduces a dynamic training objective scheduling strategy to address the issue of model training collapse during the joint optimization process when using original samples directly. 3SAT is not only structurally simple and computationally efficient, reducing self-AT training time by half, but it also improves the SOTA self-AT robustness accuracy by 16.19\% and standard accuracy by 11.41\% under Auto-Attack on the CIFAR-10 dataset. Even more impressively, 3SAT surpasses the SOTA sup-AT method in robust accuracy by a significant margin of 11.25\%. This marks the first time that self-AT has outperformed SOTA sup-AT in robustness, indicating that self-AT is a superior method for improving model robustness.

Esports has rapidly emerged as a global phenomenon with an ever-expanding audience on livestream platforms. However, due to the complex nature of the game, it becomes challenging for newcomers to comprehend the gaming situation. This research introduces a 3M-Game that integrates multi-modal (MM) information from the livestream platform, including chat and livestream, to uncover the event. While conventional MM models typically prioritise aligning MM data through concurrent training towards a unified objective, our framework leverages multiple independent teachers trained on different tasks to accomplish game event detection. The results show the effectiveness of the proposed framework. The code and appendix are in https://github.com/adlnlp/3m_game.

Advancements in neural implicit representations and differentiable rendering have markedly improved the ability to learn animatable 3D avatars from sparse multi-view RGB videos. However, current methods that map observation space to canonical space often face challenges in capturing pose-dependent details and generalizing to novel poses. While diffusion models have demonstrated remarkable zero-shot capabilities in 2D image generation, their potential for creating animatable 3D avatars from 2D inputs remains underexplored. In this work, we introduce 3D²-Actor, a novel approach featuring a pose-conditioned 3D-aware human modeling pipeline that integrates iterative 2D denoising and 3D rectifying steps. The 2D denoiser, guided by pose cues, generates detailed multi-view images that provide the rich feature set necessary for high-fidelity 3D reconstruction and pose rendering. Complementing this, our Gaussian-based 3D rectifier renders images with enhanced 3D consistency through a two-stage projection strategy and a novel local coordinate representation. Additionally, we propose an innovative sampling strategy to ensure smooth temporal continuity across frames in video synthesis. Our method effectively addresses the limitations of traditional numerical solutions in handling ill-posed mappings, producing realistic and animatable 3D human avatars. Experimental results demonstrate that 3D²-Actor excels in high-fidelity avatar modeling and robustly generalizes to novel poses.

In this paper, we propose a new benchmark called "Archaeological Piece Grouping." In the field of archaeology, it is common for broken archaeological pieces, such as artifact fragments, to be mixed. Archaeologists often spend significant time distinguishing these pieces and categorizing them into different groups. Our benchmark introduces a novel, comprehensive dataset named ArcPie, along with new evaluation metrics for this task. Additionally, we propose a new framework called "3D Probabilistic Graph Search" (3DPGS) to address the problem of grouping mixed archaeological pieces. This framework includes a relation network designed to learn the relationships among all the input 3D pieces. Utilizing the relationships learned, our framework generates a probabilistic matching graph that describes the affinity of any two pieces. We also introduce a novel search algorithm to identify groups according to this matrix. Our framework significantly outperforms other baselines.

Noise is an inevitable aspect of point cloud acquisition, necessitating filtering as a fundamental task within the realm of 3D vision. Existing learning-based filtering methods have shown promising capabilities on commonly used datasets. Nonetheless, the effectiveness of these methods is constrained when dealing with a substantial quantity of point clouds. This limitation primarily stems from their limited denoising capabilities for dense and large-scale point clouds and their inclination to generate noisy outliers after denoising. To deal with this challenge, we introduce 3DMambaIPF, for the first time, exploiting Selective State Space Models (SSMs) architecture to handle highly-dense and large-scale point clouds, capitalizing on its strengths in selective input processing and large context modeling capabilities. Additionally, we present a robust and fast differentiable rendering loss to constrain the noisy points around the surface. In contrast to previous methodologies, this differentiable rendering loss enhances the visual realism of denoised geometric structures and aligns point cloud boundaries more closely with those observed in real-world objects. Extensive evaluations on commonly used datasets (typically with up to 50K points) demonstrate that 3DMambaIPF achieves state-of-the-art results. Moreover, we showcase the superior scalability and efficiency of 3DMambaIPF on highly dense and large-scale point clouds with up to 500K points compared to off-the-shelf methods.

The rapid advancement of 3D Generative Adversarial Networks (GANs) has significantly enhanced the diversity and quality of generated 3D images. Despite these breakthroughs, the manipulation capabilities of 3D GANs remain unexplored, presenting substantial challenges for practical applications where user interaction and modification are essential. Current manipulation methods often lack the precision needed for fine-grained attribute manipulation, and struggle to maintain multi-view consistency during the editing process. To address these limitations, we propose 3DHumanEdit, a novel approach for 3D human body part-aware manipulation. 3DHumanEdit leverages multi-modal feature fusion and body part-aware feature alignment to achieve precise manipulation of individual body parts based on detailed text inputs and segmentation images. By exploring 3D prior for accurate editing and enforcing correspondence in latent space, 3DHumanEdit ensures coherence across multiple views. Experiments demonstrate that 3DHumanEdit outperforms existing methods in both editing fidelity and multi-view consistency, offering a robust solution for fine-grained 3D manipulation.

Aerial Action Recognition (AAR) in videos captured by Unmanned Aerial Vehicles (UAVs) plays a vital role in numerous applications. However, current methods related to traditional action recognition primarily cater to fixed or near cameras, and rarely consider the movement disturbance of UAVs, including their varying attitudes and positions. Those characteristics of aerial videos bring moving objects in small regions compared to broad backgrounds and relative movement to the motion of objects, which reflect more sparse and disturbed semantic information for AAR. To address these issues, we present a novel framework, dubbed 3D-Tok, to Select, Expand, and Squeeze original visual tokens for obtaining compact yet diverse semantic-enhanced tokens. In particular, we present a 3D-token selector (3TS) to select complex yet diverse tokens in three channels, capturing the semantic awareness of moving objects in comparatively small regions. Additionally, to get rid of disturbed semantic information caused by the UAV flight, we present an Expand-Squeeze Converter (ESC) to adaptively expand and squeeze the 3D-selected tokens constrained by contrastive loss, thereby suppressing the semantic-irrelevant information and reinforce semantic-relevant information via the interpolation converting. By involving the token selecting, expanding, and squeezing into an all-in-one framework, 3D-Tok shows significant improvements on the UAV-Human dataset(↑9.5%), RoCoG-v2 dataset (↑23.5%), and Drone-Action dataset (↑5.7%).

An essential component in Large Language Models (LLMs) is Rotary Position Encoding (RoPE) , which efficiently manages positional dependencies in long-context modeling. However, when the number of input tokens surpasses the pretrained capacity of LLMs, their ability to process and generate text is markedly weakened. Although position interpolation techniques for RoPE can mitigate this issue, an increase in interpolations leads to a decrease in positional resolution. To tackle this challenge, drawing inspiration from the Bloch Sphere representation, we propose a novel rotary position encoding on a three-dimensional sphere, named 3D Rotary Position Encoding (3D-RPE). 3D-RPE is an advanced version of the widely used 2D RoPE, with two major advantages for modeling long contexts: controllable long-term decay and improved position resolution. For controllable long-term decay, 3D-RPE allows for the regulation of long-term decay within the chunk size, ensuring the modeling of relative positional information between tokens at a distant relative position. For improved position resolution, 3D-RPE can mitigate the degradation of position resolution caused by position interpolation on RoPE. We have conducted experiments on long-context Natural Language Understanding (NLU) and long sequence Language Modeling (LM) tasks. From the experimental results, 3D-RPE achieved performance improvements over RoPE, especially in long-context NLU tasks.

In structured light systems, the accuracy of measurement notably diminishes when assessing complex texture objects, especially encountering boundaries between various colors. To address this challenge, this paper meticulously analyzes and establishes an error model, elaborating the correlation between phase errors and the gradients of phase and gray-scale. Based on this analysis, a novel high-precision method is proposed for measuring complex texture objects via bidirectional fringe projection. This approach firstly leverages horizontal and vertical fringe projections to derive bidirectional phase information and calculates the angles between the tangent of the texture edges and the phase gradient. Subsequently, a refined temporal phase correction algorithm is formulated based on the epipolar matching algorithm and the devised error model, effectively mitigating numerical instability issues within the algorithm and significantly reducing errors of bidirectional phases. Ultimately, corrected point clouds are calculated based on bidirectional phases, and the obtained point clouds are merged to further diminish phase errors. Comparison experiments indicate that this method can reduce Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by 65.74% and 67.75%, respectively. Compared to existing methods, it improves performance by 27.29% and 33.74%, respectively, demonstrating superior performance.