AAAI 2025 Papers

Step-level reward models (SRMs) can significantly enhance mathematical reasoning performance through process supervision or step-level preference alignment based on reinforcement learning. The performance of SRMs is pivotal, as they serve as critical guidelines, ensuring that each step in the reasoning process is aligned with desired outcomes. Recently, AlphaZero-like methods, where Monte Carlo Tree Search (MCTS) is employed for automatic step-level preference annotation, have proven particularly effective. However, the precise mechanisms behind the success of SRMs remain largely unexplored. To address this gap, this study delves into the counterintuitive aspects of SRMs, particularly focusing on MCTS-based approaches. Our findings reveal that the removal of natural language descriptions of thought processes has minimal impact on the efficacy of SRMs. Furthermore, we demonstrate that SRMs are adept at assessing the complex logical coherence present in mathematical language while having difficulty in natural language. These insights provide a nuanced understanding of the core elements that drive effective step-level reward modeling in mathematical reasoning. By shedding light on these mechanisms, this study offers valuable guidance for developing more efficient and streamlined SRMs, which can be achieved by focusing on the crucial parts of mathematical reasoning.

Serial dictatorship is a simple mechanism for coordinating agents in solving combinatorial optimization problems according to their preferences. The most representative such problem is one-sided matching, in which a set of n agents have values for a set of n items, and the objective is to compute a matching of the agents to the items of maximum total value (a.k.a., social welfare). Following the recent framework of Caragiannis and Rathi (2023), we consider a model in which the agent-item values are not available upfront but become known by querying agent sequences. In particular, when the agents are asked to act in a sequence, they respond by picking their favorite item that has not been picked by agents who acted before and reveal their value for it. Can we compute an agent sequence that induces a social welfare-optimal matching? We answer this question affirmatively and present an algorithm that uses polynomial number (specifically, O(n^5) of queries). This solves the main open problem stated by Caragiannis and Rathi (2023). Our analysis uses a potential function argument that measures progress towards learning the underlying edge-weight information. Furthermore, the algorithm has a truthful implementation by adapting the paradigm of VCG payments.

For large-scale point cloud processing, resampling takes the important role of controlling the point number and density while keeping the geometric consistency. However, current methods cannot balance such different requirements. Particularly with large-scale point clouds, classical methods often struggle with decreased efficiency and accuracy. To address such issues, we propose a weighted Poisson-disk (WPD) resampling method to improve the usability and efficiency for the processing. We first design an initial Poisson resampling with a voxel-based estimation strategy. It is able to estimate a more accurate radius of the Poisson-disk while maintaining high efficiency. Then, we design a weighted tangent smoothing step to further optimize the Voronoi diagram for each point. At the same time, sharp features are detected and kept in the optimized results with isotropic property. Finally, we achieve a resampling copy from the original point cloud with the specified point number, uniform density, and high-quality geometric consistency. Experiments show that our method significantly improves the performance of large-scale point cloud resampling for different applications, and provides a highly practical solution.

Learning low-dimensional numerical representations from symbolic data, e.g., embedding the nodes of a graph into a geometric space, is an important concept in machine learning. While embedding into Euclidean space is common, recent observations indicate that hyperbolic geometry is better suited to represent hierarchical information and heterogeneous data (e.g., graphs with a scale-free degree distribution). Despite their potential for more accurate representations, hyperbolic embeddings also have downsides like being more difficult to compute and harder to use in downstream tasks. We propose embedding into a weighted space, which is closely related to hyperbolic geometry but mathematically simpler. We provide the embedding algorithm WEmbed and demonstrate, based on generated as well as over 2000 real-world graphs, that our weighted embeddings heavily outperform state-of-the-art Euclidean embeddings for heterogeneous graphs while using fewer dimensions. The running time of WEmbed and embedding quality for the remaining instances is on par with state-of-the-art Euclidean embedders.

LLM-based autonomous agents often fail to execute complex web tasks that require dynamic interaction, largely due to the inherent uncertainty and complexity of these environments. Existing LLM-based web agents typically rely on rigid, expert-designed policies specific to certain states and actions, lacking the flexibility and generalizability needed to adapt to unseen tasks. In contrast, humans excel by exploring unknowns, continuously adapting strategies based on new observations, and resolving ambiguities through exploration. To emulate human-like adaptability, web agents need strategic exploration and complex decision-making. Monte Carlo Tree Search (MCTS) is well-suited for this, but classical MCTS struggles with vast action spaces, unpredictable state transitions, and incomplete information in web tasks. In light of this, we develop WebPilot, a multi-agent system with a dual optimization strategy that improves MCTS to better handle complex web environments. Specifically, the Global Optimization phase involves generating a high-level plan by breaking down tasks into manageable subtasks, continuously refining this plan through reflective analysis of new observations and previous subtask attempts, thereby focusing the search process and mitigating challenges posed by vast action spaces in classical MCTS. Subsequently, the Local Optimization phase executes each subtask using a tailored MCTS designed for complex environments, effectively addressing uncertainties and managing incomplete information by iteratively refining decisions based on new observations. Experimental results on WebArena and MiniWoB++ demonstrate the effectiveness of WebPilot. Notably, on WebArena, WebPilot achieves SOTA performance with GPT-4, achieving a 93% relative increase in success rate over the concurrent tree search-based method. WebPilot advances autonomous agents, enabling more reliable decision-making in practical environments.

This paper outlines a proposal regarding the use of machine learning, specifically a long-short term model, to increase the military’s effectiveness and safety protocols. The approach is to collect data from weapons training and apply it to a model that can distinguish between weapon activities. By training the model on a dataset that consists of several common weapons activities, we hope to improve commanders' understanding of their troop's performance and readiness. The evaluation will consist of examining the loss of the model, its accuracy, and analyzing activities it frequently confused. This work will extend the current research in soldier activity recognition by introducing weapon activity recognition.

Image-level weakly supervised semantic segmentation (WSSS) reduces the dependence on high-quality data annotation, which plays a crucial role in computational pathology. Benefit from the ability to localize the objects with only binary labels, Class Activation Map (CAM) is a widely used method to initial pseudo masks. However, due to the low contrast among different tissues in histopathological images, most existing CAM-based methods perform poorly in gland segmentation. We retrospect this process and find that class consistency and semantic consistency can guide the network to effectively distinguish confusing pixels and generate fine-grained pseudo masks. Specifically, for class consistency, we propose Consistency Correlation Attention (CCA) to encourage the network to focus on the contribution of class features to semantic dependencies. For semantic consistency, we propose Multi-scale Pyramid Fusion Pooling (MPFP) to aggregate coarse-to-fine global semantic information from CAMs at multiple spatial resolutions, thus identifying class localization. Additionally, we introduce a Purified Labels Filtration (PLF) strategy during the segmentation phase to mitigate the noisy supervision signal and improve the segmentation quality of the model. Extensive experiments show that the our method achieves new state-of-the-art results on three publicly available gland datasets. Furthermore, our method demonstrates impressive domain adaptation capability, achieving satisfactory results with only a small portion of samples when faced with unseen domain data.

An important - but very demanding - property in collective decision-making is strategyproofness, which requires that voters cannot benefit from submitting insincere preferences. Gibbard (1977) has shown that only rather unattractive rules are strategyproof, even when allowing for randomization. However, Gibbard's theorem is based on a rather strong interpretation of strategyproofness, which deems a manipulation successful if it increases the voter's expected utility for at least one utility function consistent with his ordinal preferences. In this paper, we study weak strategyproofness, which deems a manipulation successful if it increases the voter's expected utility for all utility functions consistent with his ordinal preferences. We show how to systematically design attractive, weakly strategyproof social decision schemes (SDSs) and explore their limitations for both strict and weak preferences. In particular, for strict preferences, we show that there are weakly strategyproof SDSs that are either ex post efficient or Condorcet-consistent, while neither even-chance SDSs nor pairwise SDSs satisfy both properties and weak strategyproofness at the same time. By contrast, for the case of weak preferences, we discuss two sweeping impossibility results that preclude the existence of appealing weakly strategyproof SDSs.

Responsible AI (RAI) is the science and practice of ensuring the design, development, use, and oversight of AI are socially sustainable---benefiting diverse stakeholders while controlling the risks. Achieving this goal requires active engagement and participation from the broader public. This paper introduces "We are AI: Taking Control of Technology," a public education course that brings the topics of AI and RAI to the general audience in a peer-learning setting. We outline the goals behind the course's development, discuss the multi-year iterative process that shaped its creation, and summarize its content. We also discuss two offerings of "We are AI" to an active and engaged group of librarians and professional staff at New York University, highlighting successes and areas for improvement. The course materials, including a multilingual comic book series by the same name, are publicly available and can be used independently. By sharing our experience in creating and teaching "We are AI", we aim to introduce these resources to the community of AI educators, researchers, and practitioners, supporting their public education efforts.

Time Series Forecasting (TSF) aims at predicting future values for a time series data and plays a crucial role in many real-world applications, e.g., finance, disease spread, or weather predictions. However, it is also a very challenging task due to complex temporal dependencies in the data, especially for long-term forecasting. In this paper, we introduce WaveletMixer, an iterative multi-levels, multi-resolutions and multi-phases approach to effectively capture long-term dependencies of multivariate time series in both global and local perspectives for improving forecasting performance. WaveletMixer fundamentally differs from existing works in the following key aspects. First, it exploits multi-levels properties of Wavelet transformation to create multiple forecasting models for different frequency domains at various levels of resolutions. Second, the relationships among different frequency domains are exploited to iteratively adjust all prediction models at all levels simultaneously in both local and global perspectives to reduce prediction errors and biases, thus significantly improving the final accuracy. Third, while WaveletMixer is a general framework that can be used to boost the performance of any deep-learning architecture (e.g., MLP, LSTM or Transformer), we additionally introduce TS-Learner, an MLP-based model to further enhance the performance in long-term forecasting. Extensive experiments have been conducted on nine real-world datasets to demonstrate the outstanding performance of WaveletMixer compared to SOTA methods and to reveal its important characteristics.

Masked Image Modeling (MIM) has garnered significant attention in self-supervised learning, thanks to its impressive capacity to learn scalable visual representations tailored for downstream tasks. However, images inherently contain abundant redundant information, leading the pixel-based MIM reconstruction process to focus excessively on finer details such as textures, thus prolonging training times unnecessarily. Addressing this challenge requires a shift towards a compact representation of features during MIM reconstruction. Frequency domain analysis provides a promising avenue for achieving compact image feature representation. In contrast to the commonly used Fourier transform, wavelet transform not only offers frequency information but also preserves spatial characteristics and multi-level features of the image. Additionally, the multi-level decomposition process of wavelet transformation aligns well with the hierarchical architecture of modern neural networks. In this study, we leverage wavelet transform as a tool for efficient representation learning to expedite the training process of MIM. Specifically, we conduct multi-level decomposition of images using wavelet transform, utilizing wavelet coefficients from different levels to construct distinct reconstruction targets representing various frequencies and scales. These reconstruction targets are then integrated into the MIM process, with adjustable weights assigned to prioritize the most crucial information. Extensive experiments demonstrate that our method achieves comparable or superior performance across various downstream tasks while exhibiting higher training efficiency.

Pansharpening aims to combine a high-resolution panchromatic (PAN) image with a low-resolution multispectral (LRMS) image to produce a high-resolution multispectral (HRMS) image. Although pansharpening in the frequency domain offers clear advantages, most existing methods either continue to operate solely in the spatial domain or fail to fully exploit the benefits of the frequency domain. To address this issue, we innovatively propose Multi-Frequency Fusion Attention (MFFA), which leverages wavelet transforms to cleanly separate frequencies and enable lossless reconstruction across different frequency domains. Then, we generate Frequency-Query, Spatial-Key, and Fusion-Value based on the physical meanings represented by different features, which enables a more effective capture of specific information in the frequency domain. Additionally, we focus on the preservation of frequency features across different operations. On a broader level, our network employs a wavelet pyramid to progressively fuse information across multiple scales. Compared to previous frequency domain approaches, our network better prevents confusion and loss of different frequency features during the fusion process. Quantitative and qualitative experiments on multiple datasets demonstrate that our method outperforms existing approaches and shows significant generalization capabilities for real-world scenarios.

Recent advancements in single image super-resolution have been predominantly driven by token mixers and transformer architectures. WaveMixSR utilized the WaveMix architecture, employing a two-dimensional discrete wavelet transform for spatial token mixing, achieving superior performance in super-resolution tasks with remarkable resource efficiency. In this work, we present an enhanced version of the WaveMixSR architecture by (1) replacing the traditional transpose convolution layer with a pixel shuffle operation and (2) implementing a multistage design for higher resolution tasks (4x). Our experiments demonstrate that our enhanced model -- WaveMixSR-V2 -- outperforms other architectures in multiple super-resolution tasks, achieving state-of-the-art for the BSD100 dataset, while also consuming fewer resources and exhibiting higher parameter efficiency and throughput.

Designing an appropriate loss function can enhance the discriminative power on gait recognition. However, previous research focuses on improving network structure and enriching input modalities but overlooks the loss functions. Although transferring loss functions from face recognition can address sample-level loss, additional design is needed for part-level loss. Therefore, we have designed a new loss function called Waveloss, aimed at adaptively and dynamically changing the preference for parts of different difficulties. First, the previous method treats the loss of different parts equally, which brings the problems of difficult convergence or susceptibility to noise interference, so we propose norm-fusion to adaptively learn samples of different difficulties. Additionally, since we find the exponential value represents preference for learning different samples, we introduce the Dynamic Learning Process, which dynamically adjusts the exponential value during iteration to focus on samples of varying difficulties at different training stages. Finally, as the changes of the exponential value leads to significant fluctuations in the gradient, we introduce the gradient truncation and normalization to avoid getting trapped in local optima and gradient vanishing or exploding by adaptively adjusting the gradient. Experimental results demonstrate that our proposed Waveloss achieves state-of-the-art performance on various gait recognition datasets and can improve the performance of different backbones as well.

Underwater salient object detection (USOD) plays a pivotal role in various vision-based marine exploration tasks. However, existing USOD techniques face the dilemma of object mislocalization and imprecise boundaries due to the complex underwater environment. The quality degradation of raw underwater images (caused by selective absorption and medium scattering) makes it challenging to perform instance detection directly. One conceivable approach involves initially removing visual disturbances through underwater image enhancement (UIE), followed by saliency detection. However, this two-stage approach neglects the potential positive impact of the restoration procedure on saliency detection due to it executes in a cascade. Based on this insight, we propose a generalized prior-involved diffusion model, called WaterDiffusion for collaborative underwater saliency detection and visual restoration. Specifically, we first propose a revised self-attention joint diffusion, which embeds dynamic saliency masks into the diffusive network as latent features. By extending the underwater degradation prior into the multi-scale decoder, we innovatively exploit optical transmission maps to aid in localizing underwater salient objects. Then, we further design a gate-guided binary indicator to select either normalized or raw channels for improving feature generalization. Finally, the Half-quadratic Splitting is introduced into the unfolding sampling to refine saliency masks iteratively. Comprehensive experiments demonstrate the superior performance of WaterDiffusion over state-of-the-art methods in both quantitative and qualitative evaluations.

The goal of video moment retrieval and highlight detection is to identify specific segments and highlights based on a given text query. With the rapid growth of video content and the overlap between these tasks, recent works have addressed both simultaneously. However, they still struggle to fully capture the overall video context, making it challenging to determine which words are most relevant. In this paper, we present a novel Video Context-aware Keyword Attention module that overcomes this limitation by capturing keyword variation within the context of the entire video. To achieve this, we introduce a video context clustering module that provides concise representations of the overall video context, thereby enhancing the understanding of keyword dynamics. Furthermore, we propose a keyword weight detection module with keyword-aware contrastive learning that incorporates keyword information to enhance fine-grained alignment between visual and textual features. Extensive experiments on the QVHighlights, TVSum, and Charades-STA benchmarks demonstrate that our proposed method significantly improves performance in moment retrieval and highlight detection tasks compared to existing approaches.

Mainstream backdoor attacks on large language models (LLMs) typically set a fixed trigger in the input instance and specific responses for triggered queries. However, the fixed trigger setting (e.g., unusual words) may be easily detected by human detection, limiting the effectiveness and practicality in real-world scenarios. To enhance the stealthiness of backdoor activation, we present a new poisoning paradigm against LLMs triggered by specifying generation conditions, which are commonly adopted strategies by users during model inference. The poisoned model performs normally for output under normal/other generation conditions, while becomes harmful for output under target generation conditions. To achieve this objective, we introduce BrieFool, an efficient attack framework. It leverages the characteristics of generation conditions by efficient instruction sampling and poisoning data generation, thereby influencing the behavior of LLMs under target conditions. Our attack can be generally divided into two types with different targets: Safety unalignment attack and Ability degradation attack. Our extensive experiments demonstrate that BrieFool is effective across safety domains and ability domains, achieving higher success rates than baseline methods, with 94.3% on GPT-3.5-turbo.

As a fundamental problem of graph analysis, graph visualization aims to embed a set of graphs in a low-dimensional (e.g., 2D) space and provide insights into their distribution and clustering structure. Focusing on this problem, we propose a novel Wasserstein t-distributed embedding (WatE) method, leading to an information-enriched graph visualization paradigm. Our method learns a graph neural network to represent each graph as the mean and covariance of its node embedding distribution. Accordingly, our method can visualize each graph as an ellipse (determined by the mean and the covariance) rather than a single point. The positions of different ellipses reveal the relations among different graphs as traditional visualization methods do, while the size and shape of an ellipse preserve the node-level structural information of the corresponding graph. We propose a regularized t-distributed stochastic neighbor embedding (Rt-SNE) framework to learn the visualization model, deriving a Wasserstein distance-based Student's t-distribution of graph pairs and fitting the distribution to the data distribution under regularization. Both subjective and objective evaluations demonstrate that WatE achieves encouraging performance in various graph visualization and clustering tasks.

Model knowledge editing has become a widely researched topic because it enables efficient and rapid injection of new knowledge into language models or the correction of erroneous or outdated knowledge. Existing model knowledge editing methods typically categorized into single-instance sequential editing and massive one-time editing. However, in practical applications, the batched and iterative editing manner better aligns with model updating patterns. In this work, we explored the performance of parameter-update-based models in a new batched iterative editing benchmark. Our findings show that with an increase in the number of editing iterations, the accumulation of updated parameters leads to a greater change in the distribution of model parameters, making it more challenging to maintain editing performance and model stability. To address this degradation issue, we propose two methods: the Wasserstein distance constraint and update parameter sparsification, where the Wasserstein distance constraint optimizes the transition of parameter distribution before and after the editing, and update parameter sparsification significantly reduces the number of update parameters, thereby alleviating the issue of instability in the parameter distribution caused by the accumulation of update parameters through iterations. Our methods can be generally applied to different parameter-update-based knowledge editing models. Experiments on the zsRE and CounterFact datasets demonstrate that our methods can improve editing performance and enhance the later-stage stability of batched iterative editing across different models.