AAAI 2025 论文列表

AI for public sector research is about using AI to tackle the numerous challenges faced by public sector organizations when they are out there making our world a better place. AI for public sector research is use-inspired research. It differs from traditional AI research first and foremost in its key objective being measurable societal impact. AI for public sector research contributes to the computing community by proposing new problem models, raising complexities that challenge abstractions which often leads to new methodologies, and introducing new contexts for evaluation. However, fulfilling this promise is easier said than done. This talk consists of three parts about our preliminary work in our long-term quest to make AI for public sector operationally scalable, financially sustainable, technically generalizable, and socially responsible. We will cover (1) a concrete AI for public sector project from problem scoping to field trials and deployment, (2) a generalizable algorithm applicable to various public sector domains, and (3) an overview of our work in a wide variety of applications.

To build a responsible data economy and protect data ownerhip, it is crucial to enable learning models from separate, heterogeneous data sources without centralization. For example, federated learning (FL) aims to train models across massive remote devices or isolated organizations, while keeping user data local. However, federated learning can face critical practical issues such as scalability, noisy samples, biased learning systems or procedures, and privacy leakage. At the intersection between optimization, trustworthy (fair, robust, and private) ML, and learning in heterogeneous environments, my research aims to support scalable and responsible data sharing to collectively build intelligent models.

Probabilistic inference is a fundamental challenge in machine learning, spanning tasks from approximate Bayesian inference to generative AI. In this talk, I will present theoretically-guaranteed scalable and efficient probabilistic inference with applications in Bayesian deep learning and generative modeling. First, I will introduce a new compute paradigm for probabilistic inference that leverages modern accelerators, specifically low-precision and sparsity, to significantly speed up inference while preserving accuracy. Next, I will present a new framework for efficient inference in discrete domains, utilizing gradient information—a largely overlooked feature of discrete distributions—to enable more informed and directional exploration. Finally, I will showcase experimental results demonstrating the effectiveness of these methods across various ML tasks, including Bayesian neural networks, energy-based models, and large language models.

Recent advances in vision-language models have shown remarkable potential, yet creating scalable systems that can effectively understand and generate across modalities remains challenging. This talk will present our contributions to advancing scalable vision-language systems, focusing on three key themes: (1) efficient vision-language understanding, including our work on temporal perceiving video-language pre-training and knowledge-enhanced zero-shot retrieval; (2) scalable generation frameworks, encompassing our innovations in zero-shot captioning and co-speech gesture generation; and (3) practical applications and deployments of these technologies. We will discuss how these advances have enabled both better performance and improved efficiency in real-world scenarios, and explore future directions for scalable multimodal systems.

Trajectory-user linking (TUL) aims to match anonymous trajectories to the most likely users who generated them, offering benefits for a wide range of real-world spatio-temporal applications. However, existing TUL methods are limited by high model complexity and poor learning of the effective representations of trajectories, rendering them ineffective in handling large-scale user trajectory data.In this work, we propose a novel Scalable Trajectory-User Linking with dual-stream representation networks for large-scale TUL problem, named ScaleTUL Specifically, ScaleTUL generates two views using temporal and spatial augmentations to exploit supervised contrastive learning framework to effectively capture the irregularities of trajectories. In each view, a dual-stream trajectory encoder consisting of a long-term encoder and a short-term encoder is designed to learn the unified representations of trajectories that fuses different temporal-spatial dependencies. Then, a TUL layer is used to associate the trajectories with the corresponding users in the representation space using a two-stage training model.Experimental results on check-in mobility datasets from three real-world cities and the nationwide U.S. demonstrate the superiority of ScaleTUL over state-of-the-art baselines for large-scale TUL tasks.

Verifying safety of neural network control systems that use images as input is a difficult problem because, from a given system state, there is no known way to mathematically model what images are possible in the real-world. We build upon recent work that considers a surrogate verification approach, training a conditional generative adversarial network (cGAN) as an image generator in place of the real world. This setup enables set-based formal analysis of the closed-loop system, providing analysis beyond simulation and testing. While existing work is effective on small examples, excessive overapproximation both within a single control period (one-step error) and across multiple periods (multi-step error) limits its scalability. We propose approaches to overcome these errors. First, we address one-step error by composing the system's dynamics along with the cGAN and neural network controller, without losing the dependencies between input states and the control outputs as in the monotonic analysis of the system dynamics. Second, we reduce multi-step error by repeating the single-step composition, essentially unrolling multiple steps of the control loop into a large neural network. We then leverage existing network verification algorithms to compute accurate reachable sets for multiple steps, avoiding the accumulation of abstraction error at each step.We demonstrate the effectiveness of our approach in terms of both accuracy and scalability using two case studies. On the aircraft taxiing system, the converged reachable set is 175% larger using the prior baseline method compared with our proposed approach. On the emergency braking system, with 24x the number of image output variables from the cGAN, the baseline method fails to prove any states are safe, whereas our improvements enable set-based safety analysis.

Despite significant advancements in solving Markov Decision Processes (MDPs) and Simple Stochastic Games (SGs), scalability remains a challenge due to the exponential growth of their state spaces. This thesis aims to push the boundaries of state-of-the-art methods by tackling this issue using 1) explainability and 2) exploiting the model structure. First, we introduce the *1-2-3-Go* approach, which learns explainable policies from small MDP models and generalizes them to larger instances, improving scalability in MDPs. We then extend *Optimistic Value Iteration (OVI)* and *Sound Value Iteration (SVI)*—originally designed for MDPs—to SGs, improving efficiency in adversarial settings. Finally, we aim to exploit the *explainable policy representations* and the *model structure* to enhance both scalability and interpretability in SGs. This thesis contributes to both theoretical advancements and practical solutions for decision-making systems under uncertainty.

Hard combinatorial optimization problems, often mapped to Ising models, promise potential solutions with quantum advantage but are constrained by limited qubit counts in near-term devices. We present an innovative quantum-inspired framework that dynamically compresses large Ising models to fit available quantum hardware of different sizes. Thus, we aim to bridge the gap between large-scale optimization and current hardware capabilities. Our method leverages a physics-inspired GNN architecture to capture complex interactions in Ising models and accurately predict alignments among neighboring spins (aka qubits) at ground states. By progressively merging such aligned spins, we can reduce the model size while preserving the underlying optimization structure. It also provides a natural trade-off between the solution quality and size reduction, meeting different hardware constraints of quantum computing devices. Extensive numerical studies on Ising instances of diverse topologies show that our method can reduce instance size at multiple levels with virtually no losses in solution quality on the latest D-wave quantum annealers.

Multi-view clustering has gained increasing attention by utilizing the complementary and consensus information across views. To alleviate the computation cost for the existing multi-view clustering approaches on datasets with large scales, studies based on anchor have been presented. Although extensively adopted in the real scenarios, most of these works ignore to learn an integral subspace revealing the cluster structure with anchors from different views being aligned, where the centroid and cluster assignment matrix can be directly achieved based on the integral subspace. Moreover, these works neglect to perform the alignment among anchors and integral subspace learning in a unified model on the incomplete multi-view dataset. Then the mutual improvements among aligning anchors and learning integral subspace are not guaranteed in optimizing the objective function, which inevitably limit the representation ability of the model and result in the suboptimal clustering performance. In this paper, we propose a novel anchor learning method for incomplete multi-view dataset termed Scalable One-pass incomplete Multi-view clustEring by Aligning anchorS (SOME-AS). Specifically, we capture the complementary information among multiple views by building the anchor graph for each view on the incomplete dataset. The integral subspace reflecting the cluster structure is learned with the alignment among anchors from different views being considered. We build the cluster assignment and centroid representation with orthogonal constraint to approximate the integral subspace. Then the subspace itself and the partition are simultaneously taken into account in this manner. Besides, the mutual improvements among aligning anchors and learning integral subspace are able to be ensured. Experiments on several incomplete multi-view datasets validate the efficiency and effectiveness of SOME-AS.

Knowledge refactoring compresses logic programs by replacing them with new rules. Current approaches struggle to scale to large programs. To overcome this limitation, we introduce a constrained optimisation refactoring approach. Our first key idea is to encode the problem with decision variables based on literals rather than rules. Our second key idea is to focus on linear invented rules. Our empirical results on multiple domains show that our approach can refactor programs quicker and with more compression than the previous state-of-the-art approach, sometimes by 60%.

Artificial Intelligence (AI) continues to evolve rapidly, impacting numerous fields, including time series (TS) classification and human activity recognition (HAR). Despite the advancements in deep learning models, such as Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), these models face several challenges, including the need for extensive labeled datasets, significant computational resources, and lack of interpretability. This research aims to address these limitations by developing an adaptive hierarchical deep neural network framework that integrates fuzzy logic principles and adaptive learning techniques for robust, computationally efficient, and interpretable real-time TS analysis. The reduction in the number of parameters and the efficient learning of hierarchical features mean that less training data is needed to achieve robust performance. The model's ability to generalize from hierarchical representations allows it to make effective use of smaller datasets, which is particularly advantageous in scenarios where data is limited or expensive to obtain.The proposed framework specifically targets HAR applications using data from wearable sensors.

Multi-view clustering (MVC) methods have garnered considerable attention within centralized data frameworks. However, real-world multi-view data are often collected and stored by different organizations, complicating the practical deployment of MVC and motivating the emergence of federated multi-view clustering (FMVC). Existing FMVC approaches typically necessitate post-processing to derive clustering labels and confront challenges in effectively exploring the complementary and consistent information across multi-view data residing in different entities. To address these limitations, we propose a novel framework termed Scalable Federated One-Step Multi-View Clustering with Tensorized Regularization (SFOMVC-TR). This framework facilitates one-step clustering at each client and employs tensor learning to capture consistent and complementary information through a centralized server. Additionally, it adopts anchor graphs to enhance clustering efficiency and scalability in high-dimensional data. By incorporating a Lp,q sparse regularization on the projection matrix, SFOMVC-TR enables the direct projection of anchors into clustering assignments to mitigate redundancy. A federated optimization framework is developed to support collaborative and privacy-preserving training under the coordination of the server. Extensive experiments on multiple datasets validate the privacy and effectiveness of our method.

In numerous settings, agents lack sufficient data to learn a model directly. Collaborating with other agents may help, but introduces a bias-variance trade-off when local data distributions differ. A key challenge is for each agent to identify clients with similar distributions while learning the model, a problem that remains largely unresolved. This study focuses on a particular instance of the overarching problem, where each agent collects samples from a real-valued distribution over time to estimate its mean. Existing algorithms face impractical per-agent space and time complexities (linear in the number of agents |A|). To address scalability challenges, we propose a framework where agents self-organize into a graph, allowing each agent to communicate with only a selected number of peers r. We propose two collaborative mean estimation algorithms: one employs a consensus-based approach, while the other uses a message-passing scheme, with complexity O(r) and O(r log |A|), respectively. We establish conditions for both algorithms to yield asymptotically optimal estimates and we provide a theoretical characterization of their performance.

Derivative-free optimization algorithms play an important role in scientific and engineering design optimization problems, especially when derivative information is not accessible. In this paper, we study the framework of sequential classification-based derivative-free optimization algorithms. By introducing learning theoretic concept hypothesis-target shattering rate, we revisit the computational complexity upper bound of SRACOS Inspired by the revisited upper bound, we propose an algorithm named RACE-CARS, which adds a random region-shrinking step compared with SRACOS. We further establish theorems showing the acceleration by region shrinking. Experiments on the synthetic functions as well as black-box tuning for language-model-as-a-service demonstrate empirically the efficiency of RACE-CARS. An ablation experiment on the introduced hyper-parameters is also conducted, revealing the mechanism of RACE-CARS and putting forward an empirical hyperparameter tuning guidance.

Byte Pair Encoding (BPE) serves as a foundation method for text tokenization in the Natural Language Processing (NLP) field. Despite its wide adoption, the original BPE algorithm harbors an inherent flaw: it inadvertently introduces a frequency imbalance for tokens in the text corpus. Since BPE iteratively merges the most frequent token pair in the text corpus to generate a new token and keeps all generated tokens in the vocabulary, it unavoidably holds tokens that primarily act as components of a longer token and appear infrequently on their own. We term such tokens as Scaffold Tokens. Due to their infrequent occurrences in the text corpus, Scaffold Tokens pose a learning imbalance issue. To address that issue, we propose Scaffold-BPE, which incorporates a dynamic scaffold token removal mechanism by parameter-free, computation-light, and easy-to-implement modifications to the original BPE method. This novel approach ensures the exclusion of low-frequency Scaffold Tokens from the token representations for given texts, thereby mitigating the issue of frequency imbalance and facilitating model training. On extensive experiments across language modeling and even machine translation, Scaffold-BPE consistently outperforms the original BPE, well demonstrating its effectiveness.

Geographic information is essential for modeling tasks in fields ranging from ecology to epidemiology. However, extracting relevant location characteristics for a given task can be challenging, often requiring expensive data fusion or distillation from massive global imagery datasets. To address this challenge, we introduce Satellite Contrastive Location-Image Pretraining (SatCLIP). This global, general-purpose geographic location encoder learns an implicit representation of locations by matching CNN and ViT inferred visual patterns of openly available satellite imagery with their geographic coordinates. The resulting SatCLIP location encoder efficiently summarizes the characteristics of any given location for convenient use in downstream tasks. In our experiments, we use SatCLIP embeddings to improve performance on nine diverse geospatial prediction tasks including temperature prediction, animal recognition, and population density estimation. Across tasks, SatCLIP consistently outperforms alternative location encoders and shows promise for improving geographic domain adaptation. These results demonstrate the potential of vision-location models to learn meaningful representations of our planet from the vast, varied, and largely untapped modalities of geospatial data.

Large language models (LLMs) have achieved outstanding performance in natural language processing, but enormous model sizes and high computational costs limit their practical deployment. Structured pruning can effectively reduce the resource demands for deployment by removing redundant model parameters. However, the randomly selected calibration data and fixed single importance estimation metrics in existing structured pruning methods lead to degraded performance of pruned models. This study introduces AdaPruner, a sample-aware adaptive structured pruning framework for LLMs, aiming to optimize the calibration data and importance estimation metrics in the structured pruning process. Specifically, AdaPruner effectively removes redundant parameters from LLMs by constructing a structured pruning solution space and then employing Bayesian optimization to adaptively search for the optimal calibration data and importance estimation metrics. Experimental results show that the AdaPruner outperforms existing structured pruning methods on a family of LLMs with varying pruning ratios, demonstrating its applicability and robustness. Remarkably, at a 20% pruning ratio, the model pruned with AdaPruner maintains 97% of the performance of the unpruned model.

Consider public health officials aiming to spread awareness about a new vaccine in a community interconnected by a social network. How can they distribute information with minimal resources, so as to avoid polarization and ensure community-wide convergence of opinion? To tackle such challenges, we initiate the study of sample complexity of opinion formation in networks. Our framework is built on the recognized opinion formation game, where we regard each agent’s opinion as a data-derived model, unlike previous works that treat opinions as data-independent scalars. The opinion model for every agent is initially learned from its local samples and evolves game-theoretically as all agents communicate with neighbors and revise their models towards an equilibrium. Our focus is on the sample complexity needed to ensure that the opinions converge to an equilibrium such that every agent’s final model has low generalization error. Our paper has two main technical results. First, we present a novel polynomial time optimization framework to quantify the total sample complexity for arbitrary networks, when the underlying learning problem is (generalized) linear regression. Second, we leverage this optimization to study the network gain which measures the improvement of sample complexity when learning over a network compared to that in isolation. Towards this end, we derive network gain bounds for various network classes including cliques, star graphs, and random regular graphs. Additionally, our framework provides a method to study sample distribution within the network, suggesting that it is sufficient to allocate samples inversely to the degree. Empirical results on both synthetic and real-world networks strongly support our theoretical findings.

Online Class-Incremental Learning (OCIL) enables a model to learn new classes from a data stream. Since data stream samples are seen only once and the capacity of storage is constrained, OCIL is particularly susceptible to Catastrophic Forgetting (CF). While exemplar replay methods alleviate CF by storing representative samples, the limited capacity of the buffer inhibits capturing the entire old data distribution, leading to CF. In this regard, recent papers suggest image compression for better memory usage. However, existing methods raise two concerns: computational overhead and compression defects. On one hand, computational overhead can limit their applicability in OCIL settings, as models might miss learning opportunities from the current streaming data if computational resources are budgeted and preoccupied with compression. On the other hand, typical compression schemes demanding low computational overhead, such as JPEG, introduce noise detrimental to training. To address these issues, we propose Salient Frequency-aware Exemplar Compression (SFEC), an efficient and effective JPEG-based compression framework. SFEC exploits saliency information in the frequency domain to reduce negative impacts from compression artifacts for learning. Moreover, SFEC employs weighted sampling for exemplar elimination based on the distance between raw and compressed data to mitigate artifacts further. Our experiments employing the baseline OCIL method on benchmark datasets such as CIFAR-100 and Mini-ImageNet demonstrate the superiority of SFEC over previous exemplar compression methods in streaming scenarios.