Recently, Large Language Model (LLM)-empowered recommender systems (RecSys) have brought significant advances in personalized user experience and have attracted considerable attention. Despite the impressive progress, the research question regarding the safety vulnerability of LLM-empowered RecSys still remains largely under-investigated. Given the security and privacy concerns, it is more practical to focus on attacking the black-box RecSys, where attackers can only observe the system's inputs and outputs. However, traditional attack approaches employing reinforcement learning (RL) agents are not effective for attacking LLM-empowered RecSys due to the limited capabilities in processing complex textual inputs, planning, and reasoning. On the other hand, LLMs provide unprecedented opportunities to serve as attack agents to attack RecSys because of their impressive capability in simulating human-like decision-making processes. Therefore, in this paper, we propose a novel attack framework called CheatAgent by harnessing the human-like capabilities of LLMs, where an LLM-based agent is developed to attack LLM-Empowered RecSys. Specifically, our method first identifies the insertion position for maximum impact with minimal input modification. After that, the LLM agent is designed to generate adversarial perturbations to insert at target positions. To further improve the quality of generated perturbations, we utilize the prompt tuning technique to improve attacking strategies via feedback from the victim RecSys iteratively. Extensive experiments across three real-world datasets demonstrate the effectiveness of our proposed attacking method.
论文检索
输入标题、作者或关键词，从 642 篇学术成果中精准定位
Improving the Consistency in Cross-Lingual Cross-Modal Retrieval with 1-to-K Contrastive Learning
PDF ↗Cross-lingual Cross-modal Retrieval (CCR) is an essential task in web search, which aims to break the barriers between modality and language simultaneously and achieves image-text retrieval in the multi-lingual scenario with a single model. In recent years, excellent progress has been made based on cross-lingual cross-modal pre-training; particularly, the methods based on contrastive learning on large-scale data have significantly improved retrieval tasks. However, these methods directly follow the existing pre-training methods in the cross-lingual or cross-modal domain, leading to two problems of inconsistency in CCR: The methods with cross-lingual style suffer from the intra-modal error propagation, resulting in inconsistent recall performance across languages in the whole dataset. The methods with cross-modal style suffer from the inter-modal optimization direction bias, resulting in inconsistent rank across languages within each instance, which cannot be reflected by Recall@K. To solve these problems, we propose a simple but effective 1-to-K contrastive learning method, which treats each language equally and eliminates error propagation and optimization bias. In addition, we propose a new evaluation metric, Mean Rank Variance (MRV), to reflect the rank inconsistency across languages within each instance. Extensive experiments on four CCR datasets show that our method improves both recall rates and MRV with smaller-scale pre-trained data, achieving the new state-of-art.
Missing data is a pervasive issue in both scientific and engineering tasks, especially for the modeling of spatiotemporal data. Existing imputation solutions mainly include low-rank models and deep learning models. The former assumes general structural priors but has limited model capacity. The latter possesses salient expressivity, but lacks prior knowledge of the underlying spatiotemporal structures. Leveraging the strengths of both two paradigms, we demonstrate a low rankness-induced Transformer to achieve a balance between strong inductive bias and high expressivity. The exploitation of the inherent structures of spatiotemporal data enables our model to learn balanced signal-noise representations, making it generalizable for a variety of imputation tasks. We demonstrate its superiority in terms of accuracy, efficiency, and versatility in heterogeneous datasets, including traffic flow, solar energy, smart meters, and air quality. Promising empirical results provide strong conviction that incorporating time series primitives, such as low-rankness, can substantially facilitate the development of a generalizable model to approach a wide range of spatiotemporal imputation problems.
The popularity of decentralized finance has drawn attention to liquidity mining (LM). In LM, a user deposits her cryptocurrencies into liquidity pools to provide liquidity for exchanges and earn yields. Different liquidity pools offer varying yields and require different pairs of cryptocurrencies. A user can exchange a cryptocurrency for another with some exchange costs. Thus, an LM solution consists of exchange transactions and deposit transactions, guaranteeing (1) each exchange transaction must exchange one cryptocurrency for another at a specific rate (i.e., the exchange constraint); (2) the amounts of cryptocurrencies deposited in a liquidity pool must exceed the required threshold (i.e., the minimum constraint); (3) each deposit transaction must deposit a specific pair of cryptocurrencies at a certain rate in a liquidity pool (i.e., the deposit constraint); and (4) the cryptocurrencies used in the solution do not exceed the cryptocurrencies that the user has (i.e., the budget constraint). Selecting the most profitable LM solution is challenging due to the vast number of candidate solutions. To address this challenge, we define the yield maximization liquidity mining (YMLM) problem. Given a set of liquidity pools, a set of the user's cryptocurrencies, a set of exchange rates, and an evaluation function, YMLM aims to find an LM solution with maximal yields, satisfying the minimum, exchange, deposit, and budget constraints. We prove that YMLM is NP-hard and cannot be solved by algorithms with constant approximation ratios. To tackle YMLM, we propose two algorithms, namely YMLM\_GD and YMLM\_SK, with parameterized approximation ratios. Extensive experiments on both real and synthetic datasets show that our approaches outperform the baselines in yields.
The intrinsic predictability of a given time series indicates how well an (ideal) algorithm could potentially predict it when trained on the time series data. Being able to compute the intrinsic predictability helps the developers of prediction algorithms immensely in deciding whether there is further optimization potential, as it tells them how close they are to what is (theoretically) achievable. We call the intrinsic predictability the predictability upper bound ¶imax and propose a novel method for quantifying and estimating it for univariate numeric time series. So far, this has only been done for symbolic time series, even though most real-world time series are numeric by nature. We base our technique on the close relationship between entropy and predictability, utilizing the entropy rate of a time series to compute ¶imax . Since existing entropy rate estimators, such as those based on the Lempel-Ziv compression algorithm, only work for symbolic data, we develop new estimators using tolerance thresholds for matching numeric values. We demonstrate that ¶imax is an effective upper bound that characterizes the intrinsic predictability of a time series. We give formal proofs and we validate our arguments experimentally by comparing ¶imax with the prediction accuracy of different state-of-the-art models on various real-world datasets from different domains.
In numerous real-world domains, spanning from environmental monitoring to long-term medical studies, observations do not arrive in a single batch but rather over time in episodes. This challenges the traditional assumption in causal discovery of a single, observational dataset, not only because each episode may be a biased sample of the population but also because multiple episodes could differ in the causal interactions underlying the observed variables. We address these issues using notions of context switches and episodic selection bias, and introduce a framework for causal modeling of episodic data. We show under which conditions we can apply information-theoretic scoring criteria for causal discovery while preserving consistency. To in practice discover the causal model progressively over time, we propose the CONTINENT algorithm which, taking inspiration from continual learning, discovers the causal model in an online fashion without having to re-learn the model upon arrival of each new episode. Our experiments over a variety of settings including selection bias, unknown interventions, and network changes showcase that CONTINENT works well in practice and outperforms the baselines by a clear margin.
Empirically-determined scaling laws have been broadly successful in predicting the evolution of large machine learning models with training data and number of parameters. As a consequence, they have been useful for optimizing the allocation of limited resources, most notably compute time. In certain applications, storage space is an important constraint, and data format needs to be chosen carefully as a consequence. Computer vision is a prominent example: images are inherently analog, but are always stored in a digital format using a finite number of bits. Given a dataset of digital images, the number of bits L to store each of them can be further reduced using lossy data compression. This, however, can degrade the quality of the model trained on such images, since each example has lower resolution. In order to capture this trade-off and optimize storage of training data, we propose a 'storage scaling law' that describes the joint evolution of test error with sample size and number of bits per image. We prove that this law holds within a stylized model for image compression, and verify it empirically on two computer vision tasks, extracting the relevant parameters. We then show that this law can be used to optimize the lossy compression level. At given storage, models trained on optimally compressed images present a significantly smaller test error with respect to models trained on the original data. Finally, we investigate the potential benefits of randomizing the compression level.
Social networks represent complex ecosystems where the interactions between users or groups play a pivotal role in information dissemination, opinion formation, and social interactions. Effectively harnessing event sequence data within social networks to unearth interactions among users or groups has persistently posed a challenging frontier within the realm of point processes. Current deep point process models face inherent limitations within the context of social networks, constraining both their interpretability and expressive power. These models encounter challenges in capturing interactions among users or groups and often rely on parameterized extrapolation methods when modeling intensity over non-event intervals, limiting their capacity to capture intricate intensity patterns, particularly beyond observed events. To address these challenges, this study proposes modifications to Transformer Hawkes processes (THP), leading to the development of interpretable Transformer Hawkes processes (ITHP). ITHP inherits the strengths of THP while aligning with statistical nonlinear Hawkes processes, thereby enhancing its interpretability and providing valuable insights into interactions between users or groups. Additionally, ITHP enhances the flexibility of the intensity function over non-event intervals, making it better suited to capture complex event propagation patterns in social networks. Experimental results, both on synthetic and real data, demonstrate the effectiveness of ITHP in overcoming the identified limitations. Moreover, they highlight ITHP's applicability in the context of exploring the complex impact of users or groups within social networks. Our code is available at https://github.com/waystogetthere/Interpretable-Transformer- Hawkes-Process.git.
The problem of column subset selection asks for a subset of columns from an input matrix such that the matrix can be reconstructed as accurately as possible within the span of the selected columns. A natural extension is to consider a setting where the matrix rows are partitioned into two groups, and the goal is to choose a subset of columns that minimizes the maximum reconstruction error of both groups, relative to their respective best rank-k approximation. Extending the known results of column subset selection to this fair setting is not straightforward: in certain scenarios it is unavoidable to choose columns separately for each group, resulting in double the expected column count. We propose a deterministic leverage-score sampling strategy for the fair setting and show that sampling a column subset of minimum size becomes NP-hard in the presence of two groups. Despite these negative results, we give an approximation algorithm that guarantees a solution within 1.5 times the optimal solution size. We also present practical heuristic algorithms based on rank-revealing QR factorization. Finally, we validate our methods through an extensive set of experiments using real-world data.
A point pattern is a dataset of coordinates, typically in 2D or 3D space. Point patterns are ubiquitous in diverse applications including Geographic Information Systems, Astronomy, Ecology, Biology and Medicine. Among the statistics used to quantify point patterns, most are based on Ripley's K-function, which measures the deviation of the observed pattern from a completely random arrangement of points. This approach is useful for constructing null hypothesis tests, but Ripley's K and its variants are less suitable as quantitative effect sizes because their ranges and expected values generally depend on the scale or the size of the region in which the pattern is observed. To address this, we propose a new function that behaves like a correlation coefficient for point patterns: it is tightly bounded by -1 and 1, with a value of -1 corresponding to a maximally dispersed arrangement of points, 0 indicating complete spatial randomness, and 1 representing maximal clustering. These properties are independent of scale and observation window size assuming appropriate edge correction. Evaluating our function on simulated data, we show that it has comparable statistical calibration and power to K-based baselines. We hope that the ease of interpretation of our bounded function will facilitate the analysis of spatial data across multiple fields.
Preserving individual privacy while enabling collaborative data sharing is crucial for organizations. Synthetic data generation is one solution, producing artificial data that mirrors the statistical properties of private data. While numerous techniques have been devised under differential privacy, they predominantly assume data is centralized. However, data is often distributed across multiple clients in a federated manner. In this work, we initiate the study of federated synthetic tabular data generation. Building upon a SOTA central method known as AIM, we present DistAIM and FLAIM. We first show that it is straightforward to distribute AIM, extending a recent approach based on secure multi-party computation which necessitates additional overhead, making it less suited to federated scenarios. We then demonstrate that naively federating AIM can lead to substantial degradation in utility under the presence of heterogeneity. To mitigate both issues, we propose an augmented FLAIM approach that maintains a private proxy of heterogeneity. We simulate our methods across a range of benchmark datasets under different degrees of heterogeneity and show we can improve utility while reducing overhead.
Anomalous node detection in a static graph faces significant challenges due to the rarity of anomalies and the substantial cost of labeling their deviant structure and attribute patterns. These challenges give rise to data-centric problems, including extremely imbalanced data distributions and intricate graph learning, which significantly impede machine learning and deep learning methods from discerning the patterns of graph anomalies with few labels. While these issues remain crucial, much of the current research focuses on addressing the induced technical challenges, treating the shortage of labeled data as a given. Distinct from previous efforts, this work focuses on tackling the data-centric problems by generating auxiliary training nodes that conform to the original graph topology and attribute distribution. We categorize this approach as data-centric, aiming to enhance existing anomaly detectors by training them on our synthetic data. However, the methods for generating nodes and the effectiveness of utilizing synthetic data for graph anomaly detection remain unexplored in the realm. To answer these questions, we thoroughly investigate the denoising diffusion model. Drawing from our observations on the diffusion process, we illuminate the shifts in graph energy distribution and establish two principles for designing denoising neural networks tailored to graph anomaly generation. From the insights, we propose a diffusion-based graph generation method to synthesize training nodes, which can be promptly integrated to work with existing anomaly detectors. The empirical results on eight widely-used datasets demonstrate our generated data can effectively enhance the nine state-of-the-art graph detectors' performance.
Differentiable causal discovery has made significant advancements in the learning of directed acyclic graphs. However, its application to real-world datasets remains restricted due to the ubiquity of latent confounders and the requirement to learn maximal ancestral graphs (MAGs). To date, existing differentiable MAG learning algorithms have been limited to small datasets and failed to scale to larger ones (e.g., with more than 50 variables). The key insight in this paper is that the causal skeleton, which is the undirected version of the causal graph, has potential for improving accuracy and reducing the search space of the optimization procedure, thereby enhancing the performance of differentiable causal discovery. Therefore, we seek to address a two-fold challenge to harness the potential of the causal skeleton for differentiable causal discovery in the presence of latent confounders: (1) scalable and accurate estimation of skeleton and (2) universal integration of skeleton estimation with differentiable causal discovery. To this end, we propose SPOT (Skeleton Posterior-guided OpTimization), a two-phase framework that harnesses skeleton posterior for differentiable causal discovery in the presence of latent confounders. On the contrary to a "point-estimation", SPOT seeks to estimate the posterior distribution of skeletons given the dataset. It first formulates the posterior inference as an instance of amortized inference problem and concretizes it with a supervised causal learning (SCL)-enabled solution to estimate the skeleton posterior. To incorporate the skeleton posterior with differentiable causal discovery, SPOT then features a skeleton posterior-guided stochastic optimization procedure to guide the optimization of MAGs. Extensive experiments on various datasets show that SPOT substantially outperforms SOTA methods for MAG learning. SPOT also demonstrates its effectiveness in the accuracy of skeleton posterior estimation in comparison with non-parametric bootstrap-based, or more recently, variational inference-based methods. Finally, we observe that the adoption of skeleton posterior exhibits strong promise in various causal discovery tasks.
Conventional machine learning typically assume a fixed learning objective throughout the learning process. However, for real-world tasks in open and dynamic environments, objectives can change frequently. For example, in autonomous driving, a car has several default modes, but a user's concern for speed and fuel consumption varies depending on road conditions and personal needs. We formulate this problem as learning with varied objectives (LVO), where the goal is to optimize a dynamic weighted combination of multiple sub-objectives by sequentially selecting actions that incur different losses on these sub-objectives. We propose the VaRons algorithm, which estimates the action-wise performance on each sub-objective and adaptively selects decisions according to the dynamic requirements on different sub-objectives. Further, we extend our approach to cases involving contextual representations and propose the ConVaRons algorithm, assuming parameterized linear structure that links contextual features to the main objective. Both the VaRons and ConVaRons are provably minimax optimal with respect to the time horizon T, with ConVaRons showing better dependency with the number of sub-objectives K. Experiments on dynamic classifier and real-world cluster service allocation tasks validate the effectiveness of our methods and support our theoretical findings.
Spectral Graph Neural Networks have demonstrated superior performance in graph representation learning. However, many current methods focus on employing shared polynomial coefficients for all nodes, i.e., learning node-unified filters, which limits the filters' flexibility for node-level tasks. The recent DSF attempts to overcome this limitation by learning node-wise coefficients based on positional encoding. However, the initialization and updating process of the positional encoding are burdensome, hindering scalability on large-scale graphs. In this work, we propose a scalable node-wise filter, PolyAttn. Leveraging the attention mechanism, PolyAttn can directly learn node-wise filters in an efficient manner, offering powerful representation capabilities. Building on PolyAttn, we introduce the whole model, named PolyFormer. In the lens of Graph Transformer models, PolyFormer, which calculates attention scores within nodes, shows great scalability. Moreover, the model captures spectral information, enhancing expressiveness while maintaining efficiency. With these advantages, PolyFormer offers a desirable balance between scalability and expressiveness for node-level tasks. Extensive experiments demonstrate that our proposed methods excel at learning arbitrary node-wise filters, showing superior performance on both homophilic and heterophilic graphs, and handling graphs containing up to 100 million nodes. The code is available at https://github.com/air029/PolyFormer.
The sparse dictionary coding framework represents signals as a linear combination of a few predefined dictionary atoms. It has been employed for images, time series, graph signals and recently for 2-way (or 2D) spatio-temporal data employing jointly temporal and spatial dictionaries. Large and over-complete dictionaries enable high-quality models, but also pose scalability challenges which are exacerbated in multi-dictionary settings. Hence, an important problem that we address in this paper is: How to scale multi-dictionary coding for large dictionaries and datasets? We propose a multi-dictionary atom selection technique for low-rank sparse coding named LRMDS. To enable scalability to large dictionaries and datasets, it progressively selects groups of row-column atom pairs based on their alignment with the data and performs convex relaxation coding via the corresponding sub-dictionaries. We demonstrate both theoretically and experimentally that when the data has a low-rank encoding with a sparse subset of the atoms, LRMDS is able to select them with strong guarantees under mild assumptions. Furthermore, we demonstrate the scalability and quality of LRMDS in both synthetic and real-world datasets and for a range of coding dictionaries. It achieves 3 times to 10 times speed-up compared to baselines, while obtaining up to two orders of magnitude improvement in representation quality on some of the real world datasets given a fixed target number of atoms.
Graph Prompt Learning (GPL) bridges significant disparities between pretraining and downstream applications to alleviate the knowledge transfer bottleneck in real-world graph learning. While GPL offers superior effectiveness in graph knowledge transfer and computational efficiency, the security risks posed by backdoor poisoning effects embedded in pretrained models remain largely unexplored. Our study provides a comprehensive analysis of GPL's vulnerability to backdoor attacks. We introduce CrossBA, the first cross-context backdoor attack against GPL, which manipulates only the pretraining phase without requiring knowledge of downstream applications. Our investigation reveals both theoretically and empirically that tuning trigger graphs, combined with prompt transformations, can seamlessly transfer the backdoor threat from pretrained encoders to downstream applications.Through extensive experiments involving 3 representative GPL methods across 5 distinct cross-context scenarios and 5 benchmark datasets of node and graph classification tasks, we demonstrate that CrossBA consistently achieves high attack success rates while preserving the functionality of downstream applications over clean input. We also explore potential countermeasures against CrossBA and conclude that current defenses are insufficient to mitigate CrossBA. Our study highlights the persistent backdoor threats to GPL systems, raising trustworthiness concerns in the practices of GPL techniques.
Capturing molecular knowledge with representation learning approaches holds significant potential in vast scientific fields such as chemistry and life science. An effective and generalizable molecular representation is expected to capture the consensus and complementary molecular expertise from diverse views and perspectives. However, existing works fall short in learning multi-view molecular representations, due to challenges in explicitly incorporating view information and handling molecular knowledge from heterogeneous sources. To address these issues, we present MV-Mol, a molecular representation learning model that harvests multi-view molecular expertise from chemical structures, unstructured knowledge from biomedical texts, and structured knowledge from knowledge graphs. We utilize text prompts to model view information and design a fusion architecture to extract view-based molecular representations. We develop a two-stage pre-training procedure, exploiting heterogeneous data of varying quality and quantity. Through extensive experiments, we show that MV-Mol provides improved representations that substantially benefit molecular property prediction. Additionally, MV-Mol exhibits state-of-the-art performance in multi-modal comprehension of molecular structures and texts. Code and data are available at https://github.com/PharMolix/OpenBioMed.
Fairness-aware Graph Neural Networks (GNNs) often face a challenging trade-off, where prioritizing fairness may require compromising utility. In this work, we re-examine fairness through the lens of spectral graph theory, aiming to reconcile fairness and utility within the framework of spectral graph learning. We explore the correlation between sensitive features and spectrum in GNNs, using theoretical analysis to delineate the similarity between original sensitive features and those after convolution under different spectra. Our analysis reveals a reduction in the impact of similarity when the eigenvectors associated with the largest magnitude eigenvalue exhibit directional similarity. Based on these theoretical insights, we propose FUGNN, a novel spectral graph learning approach that harmonizes the conflict between fairness and utility. FUGNN ensures algorithmic fairness and utility by truncating the spectrum and optimizing eigenvector distribution during the encoding process. The fairness-aware eigenvector selection reduces the impact of convolution on sensitive features while concurrently minimizing the sacrifice of utility. FUGNN further optimizes the distribution of eigenvectors through a transformer architecture. By incorporating the optimized spectrum into the graph convolution network, FUGNN effectively learns node representations. Experiments on six real-world datasets demonstrate the superiority of FUGNN over baseline methods. The codes are available at https://github.com/yushuowiki/FUGNN.
Graph Neural Networks (GNNs) have revolutionized graph-based machine learning, but their heavy computational demands pose challenges for latency-sensitive edge devices in practical industrial applications. In response, a new wave of methods, collectively known as GNN-to-MLP Knowledge Distillation, has emerged. They aim to transfer GNN-learned knowledge to a more efficient MLP student, which offers faster, resource-efficient inference while maintaining competitive performance compared to GNNs. However, these methods face significant challenges in situations with insufficient training data and incomplete test data, limiting their applicability in real-world applications. To address these challenges, we propose AdaGMLP, an AdaBoosting GNN-to-MLP Knowledge Distillation framework. It leverages an ensemble of diverse MLP students trained on different subsets of labeled nodes, addressing the issue of insufficient training data. Additionally, it incorporates a Node Alignment technique for robust predictions on test data with missing or incomplete features. Our experiments on seven benchmark datasets with different settings demonstrate that AdaGMLP outperforms existing G2M methods, making it suitable for a wide range of latency-sensitive real-world applications. We have submitted our code to the GitHub repository (https://github.com/WeigangLu/AdaGMLP-KDD24).