Recommendation is crucial for both user experience and company revenue in Meituan as a leading lifestyle company, and generative recommendation models (GRMs) are shown to produce quality recommendations recently. However, existing systems are limited by insufficient functionality support and inefficient implementations for training GRMs in industrial scenarios. As such, we introduce MTGenRec as an efficient and scalable system for GRM training. Specifically, to handle real-time insertions/deletions of sparse embeddings, MTGenRec employs dynamic hash tables to replace static ones. To improve training efficiency, MTGenRec conducts dynamic sequence balancing to address the computation load imbalances among GPUs and adopts feature ID deduplication alongside automatic table merging to accelerate embedding lookup. Extensive experiments show that MTGenRec improves training throughput by 1.6x - 2.4x while achieving good scalability when running over 100 GPUs. MTGenRec has been deployed for many applications in Meituan and is now handling hundreds of millions of requests on a daily basis. On the delivery platform, we observe a 1.22% growth in user order volume and a 1.31% enhancement in online PV_CTR.

Vertical multimodal federated learning (VMFL) enables multiple clients holding data from different modalities to conduct collaboratively model training. Existing methods typically assume that multimodal data samples (i.e., text and image) from the same entity (i.e., person) are paired across the clients (i.e., aligned). However, this assumption rarely holds in practice, as data is often collected independently with no shared identifiers. To address this challenge, we propose hashing-based alignment (HAL), a new VMFL framework that works without pre-aligned samples. HAL consists of two key components. The first component is an efficient and privacy-preserving method to identify similar samples from different modalities as aligned pairs. It adopts locality sensitive hashing (LSH) for the efficient retrieval of similar samples, introduces a shift-orthogonal hashing scheme to tackle the gaps between different modalities, and uses a bloom-style method for secure Hamming distance estimation. We prove that the shift-orthogonal hashing reduces distance estimation errors and secure Hamming distance estimation satisfies differential privacy. The second component is a neighbor-aware fusion strategy, which applies cross-attention to aggregate informative signals from the aligned samples without relying on explicit similarity scores. Experimental results on two real-world datasets show that compared with five state-of-the-art (SOTA) baselines, HAL improves the cross-modal retrieval accuracy by over 63%, while also achieving up to 154× speedup.

Text-Attributed Graph (TAG) is an important type of graph structured data, where each node has a text description. TAG models usually train a Graph Neural Network (GNN) and language model jointly, which leads to high space and time consumption, especially on large datasets. To mitigate this, we propose TAGSAM, a condensation method that compresses TAGs while preserving training accuracy. TAGSAM comes with two key designs, i.e., subgraph text Selection and Attribute similarity Matching, which compress the text description and graph topology of TAG, respectively. For the texts, subgraph text selection selects and merges representative text chunks from multiple related text descriptions by maximizing mutual information. For the graph topology, popular condensation methods based on Matching Training Trajectories (MTT) suffer from high variance, which hinders accuracy. Our attribute similarity matching mitigates this issue by aligning stable similarity matrices. We evaluate TAGSAM against six state-of-the-art baselines, where it showcases superior performance. For the same compressed size, TAGSAM improves upon the best-performing baseline by an average of 4.9% in accuracy. Furthermore, it maintains competitive training accuracy even when the TAG is condensed to just 1% size. Our code is available at https://github.com/SundayVHan/TAGSAM

Attribute Inference Attacks (AIAs) pose a significant threat to recommendation systems (RS) by enabling adversaries to use threat models to infer sensitive user attributes like gender or race from user embeddings, resulting in privacy breaches such as unauthorized profiling and discriminatory policies against specific groups. Existing attribute protection methods are primarily applied during training, suffering from significant limitations, such as architectural inflexibility, dependence on interaction data, and potential catastrophic degradation in recommendation performance. To overcome these challenges, we propose AttrCloak, an efficient and effective post-training attribute unlearning (AU) framework that removes sensitive information from user embeddings without altering RS training architectures. AttrCloak employs dual-objective optimization with parameter self-sharing to minimize mutual information between user embeddings and sensitive attributes while preserving recommendation quality. Furthermore, it accommodates data-free scenarios by leveraging regularization loss when interaction data is unavailable. Comprehensive evaluations on four real-world datasets demonstrate AttrCloak's good performance in privacy protection and recommendation performance.

Due to the rising concerns on privacy protection, how to build machine learning models from distributed databases with privacy guarantees has gained more popularity. Vertical federated learning (VFL) trains machine learning models in a privacy-preserving way when the data features are scattered over distributed databases. We study the participant selection problem (PSP) for VFL, which chooses a given number of participants to conduct training while maximizing model accuracy. Compared to training with all participants, PSP can filter out hitch-riders that contribute marginally to model quality and reduce training time by involving fewer participants. To achieve good model accuracy, we formulate PSP as choosing a set of participants that maximizes the likelihood of the data samples. Then, utilizing the k-nearest neighbors (KNN) classifier as the proxy model, we express the likelihood as a function of the selected participants and prove that the function is sub modular. The submodular property is favorable as it can account for the feature diversity among the participants and allows to greedily select the participant with the maximum gain in each step. However, the selection process requires finding the top-k neighbors of a data sample as the basic operation, which is expensive in VFL setting as it involves encrypted communication. As such, we adapt the Fagin's algorithm, a famous top-k query algorithm, to reduce the amount of encrypted communication. We deploy our solution VFPS-SM across five distributed nodes and conduct experiments with 10 datasets and 3 models to evaluate its performance. The results show that VFPS-SM can reduce the end-to-end running time by up to$35\times$, selection time$365\times$and improve model accuracy by 6.0% compared with state-of-the-art baselines.

Dynamic graph neural networks (DGNNs) are designed to capture the dynamic evolution of graph node interactions. However, existing DGNNs mainly consider homogeneous graphs, neglecting the rich heterogeneity in node and edge types, which is prevalent for real-world graphs and essential for modeling complex dynamic interactions. In this work, we propose the TrajEctory and Semantic-Aware dynamic heterogeneous graph neural network (TeSa), which integrates trajectory-based evolution and semantic-aware aggregation to capture both the evolving dynamics and heterogeneous semantics entailed in continuous-time dynamic heterogeneous graphs. In particular, trajectory-based evolution treats the interactions received by each node (called node trajectory) as a sequence and employs a temporal point process to learn the dynamic evolution in these interactions. Semantic-aware aggregation separates edges of different types when aggregating messages for each node from its neighbors. Edges of the same type are processed at first (i.e., intra-semantic aggregation), and then edges of different types are handled (i.e., inter-semantic fusion), to offer a comprehensive view of the heterogeneous semantics. We compare TeSa with 7 state-of-the-art DGNN models, and the results show that TeSa improves the best-performing baseline by an average of 5.11% and 5.74% in accuracy for transductive and inductive tasks.

As privacy concerns in recommender systems become increasingly prominent, federated recommender systems (FedRecs) have emerged as a promising distributed training paradigm.FedRecs enable the collaborative training of a shared global recommendation model without requiring the exchange of raw client interaction data.However, models trained using standard FedRec methods remain vulnerable to personal information leakage, particularly through attribute inference attacks, which can expose sensitive user attributes such as gender and race.In this paper, we address these user-sensitive attributes as targets for federated unlearning.To protect users' sensitive information, attribute unlearning aims to eliminate sensitive attributes from user embeddings, thereby preventing inference attacks while preserving recommendation performance.We introduce a novel post-training federated unlearning framework, Aegis, which performs unlearning based on private attribute requests after the model has been trained, minimizing the degradation in recommendation accuracy.Aegis employs an information-theoretic multi-component loss function to balance privacy protection and recommendation performance.Additionally, Aegis adapts to scenarios where training interaction data may be unavailable, reflecting real-world centralized protection scenarios.Comprehensive evaluations of various benchmark datasets demonstrate that our proposed method effectively safeguards user privacy while maintaining high-quality recommendations. CCS Concepts• Information systems → Social recommendation

Abstract Traditional fraud detection approaches often use linking entities, such as device, email, and address, to identify fraudulent transactions and users. However, as fraud methods continue to evolve and escalate, the fraudsters can fabricate the involved entities and thus hide their real intent. To make fraud detection more robust, we incorporate user behaviors in the pipeline and consider biometric characteristics that are difficult to forge. In this work, we conduct a detailed study of how user behavior data can help identify and prevent fraudulent activity in e-commerce. We present Multi-Modal Behavioral Transformer (MMBT), where we combine both inner-page behavioral data, such as mouse trajectory, and inter-page behavioral data, such as page view sequences. We propose to construct mouse trajectory data as an image, treat each mouse position as a pixel in the image, convert the image into small patches, and hence transform the mouse trajectory into patch index sequences. Our experimental results on real-word data show that MMBT significantly outperforms baselines — the precision@recall = 0.1 increases by up to 7%. In addition, we have built an online pipeline to operationalize our model. In production, the 99th percentile latency is maintained below 500 milliseconds, allowing the platform to initiate rapid response measures and prevent potential losses.

With the development of AI and the growing demand for computational power, hardware is becoming increasingly specialized and heterogeneous. The emergence of diverse specialized hardware architectures, each with distinct characteristics and programming abstractions, poses significant portability and sustainability challenges for existing data processing systems. Tensor Computation Runtimes (TCRs) abstract away the low-level hardware complexities by providing users with a hardware-independent tensor-based interface, enabling data scientists to effectively leverage the powerful capabilities of new hardware accelerators (collectively referred to as XPU). Built on TCRs, the existing relational query engine TQP demonstrates portability across a wide range of target hardware and sustainability along with the ongoing evolution of TCRs and hardware. However, it neglects the big gap between irregular SQL workloads and uniform tensor operations when mapping SQL operators to tensor programs, which causes significant storage and computation overhead. In this paper, for the first time, we analyze the underlying gap between SQL and tensors, and provide guidelines to bridge it. Following these guidelines, we build a new Tensor-based Query Engine Enhanced (TQEx) by bridging the gap from multiple aspects: develop efficient storage and computation strategies for variable-length data, and design efficient SQL operators such as join and aggregate based on tensors. We also extend TQEx to multi-XPUs for large-scale data processing. Extensive experimental studies show that our query engine, TQEx, achieves a 9.6× speedup (with a peak of 41.9×) over TQP on TPC-H, and it is also 27.9× faster than leading GPU databases such as HeavyDB. On TPC-H at scale factor 100, TQEx outperforms DuckDB by 12.2× and HeavyDB by 22.7× on supported queries.

Graph is ubiquitous in various real-world applications, and many graph processing systems have been developed. Recently, hardware accelerators have been exploited to speed up graph systems. However, such hardware-specific systems are hard to migrate across different hardware backends. In this paper, we propose the first tensor-based graph processing framework, Tgraph, which can be smoothly deployed and run on any powerful hardware accelerators (uniformly called XPU) that support Tensor Computation Runtimes (TCRs). TCRs, which are deep learning frameworks along with their runtimes and compilers, provide tensor-based interfaces to users to easily utilize specialized hardware accelerators without delving into the complex low-level programming details. However, building an efficient tensor-based graph processing framework is non-trivial. Thus, we make the following efforts: (1) propose a tensor-centric computation model for users to implement graph algorithms with easy-to-use programming interfaces; (2) provide a set of graph operators implemented by tensor to shield the computation model from the detailed tensor operators so that Tgraph can be easily migrated and deployed across different TCRs; (3) design a tensor-based graph compression and computation strategy and an out-of-XPU-memory computation strategy to handle large graphs. We conduct extensive experiments on multiple graph algorithms (BFS, WCC, SSSP, etc.), which validate that Tgraph not only outperforms seven state-of-the-art graph systems, but also can be smoothly deployed and run on multiple DL frameworks (PyTorch and TensorFlow) and hardware backends (Nvidia GPU, AMD GPU, and Apple MPS).

Vertical federated learning (VFL) trains models when multiple databases (a.k.a participants) hold different features of the same set of samples. By quantifying each participant's contribution to model training, data valuation can prevent hitch-riders and reward the instrumental parties. However, vertical federated data valuation (VFDV) is challenging because it needs to be accurate and efficient while protecting participant data privacy. In this paper, we propose a method meeting all three requirements by using projection and sampling for mutual information estimation (thus dubbed PS-MI). In particular, we first show that the utility of a participant set (a.k.a a coalition ) can be expressed as the mutual information (MI) between their features and the target labels. MI is favorable because it does not depend on the model to train (i.e., model-agnostic ) and can be estimated via k -nearest neighbor (KNN). To run KNN, instead of using costly homomorphic encryption to protect data privacy, we apply simple random projection to participant features before distance computation. We prove that random projection ensures differential privacy and preserves unbiased distance estimates. Since the contribution of a participant involves many coalitions, we adopt stratified sampling to reduce the number of coalitions while controlling estimation variance. To further improve efficiency, we incorporate optimizations including using locality sensitive hashing (LSH) to prune kNN candidates, batching kNN candidate checking for multiple coalitions, and adaptive early termination for utility evaluation. We compare PS-MI with 5 state-of-the-art VFDV methods. The results show that PS-MI yields higher accuracy and shorter running time than the baselines, and the maximum speedup can be 592×.

Outdoor billboard advertising has proven effective for commercial promotions, attracting potential customers, and boosting product sales. Auction serves as a popular method for leasing billboard usage rights, enabling a seller to rent billboards to winning users for predefined periods according to their bids. An effective auction algorithm is of great significance to maximize the efficiency of the billboard ecosystem. In contrast to a rich literature on Internet advertising auctions, well-crafted algorithms tailored for outdoor billboard auctions remain rare. In this work, we investigate the problem of outdoor billboard auctions, in the practical setting where bids are received and processed on the fly. Our goal is to maximize social welfare, namely the total benefits of auction participants, including the billboard service provider and the bidding users. To this end, we first formulate the billboard social welfare maximization problem into an Integer Linear Problem (ILP), and then reformulate the ILP into a compact form with a reduced size of constraints (at the cost of involving exponentially many primal variables), based on which we derive the dual problem. Furthermore, we design a dual oracle to handle the exponentially many dual constraints, avoiding exhaustive enumeration. We present a primal-dual online algorithm with an incentive-compatible pricing mechanism. Theoretical analysis proves the individual rationality, incentive compatibility, and computational efficiency of our online algorithm. Extensive experimental results show that the online algorithm is both effective and efficient, and achieves a good competitive ratio.

Motif detection is a graph algorithm that detects certain local structures in a graph. Although network motif has been studied in graph analytics, e.g., social network and biological network, it is yet unclear whether network motif is useful for analyzingonline transaction networkthat is generated in applications such as instant messaging and e-commerce. In an online transaction network, each vertex represents a user’s account and each edge represents a money transaction between two users. In this work, we try to analyze online transaction networks with network motifs. We design motif-based vertex embedding that integrates motif counts and centrality measurements. Furthermore, we design a distributed framework to detect motifs in large-scale online transaction networks. Our framework obtains the edge directions using a bi-directional tagging method and avoids redundant detection with a reduced view of neighboring vertices. We implement the proposed framework under the parameter server architecture. In the evaluation, we analyze different kinds of online transaction networks w.r.t the distribution of motifs and evaluate the effectiveness of motif-based embedding in downstream graph analytical tasks. The experimental results also show that our proposed motif detection framework can efficiently handle large-scale graphs.

Graph Neural Networks have demonstrated remarkable effectiveness in various graph-based tasks, but their inefficiency in training and inference poses significant challenges for scaling to real-world, large-scale applications. To address these challenges, a plethora of algorithms have been developed to accelerate GNN training and inference, garnering substantial interest from the research community. This paper presents a systematic review of these acceleration algorithms, categorizing them into three main topics: training acceleration, inference acceleration, and execution acceleration. For training acceleration, we discuss techniques like graph sampling and GNN simplification. In inference acceleration, we focus on knowledge distillation, GNN quantization, and GNN pruning. For execution acceleration, we explore GNN binarization and graph condensation. Additionally, we review several libraries related to GNN acceleration, including our Scalable Graph Learning library, and propose future research directions.

To handle graphs in which features or connections are evolving over time, a series of temporal graph neural networks (TGNNs) have been proposed. Despite the success of these TGNNs, the previous TGNN evaluations reveal several limitations regarding four critical issues: 1) inconsistent datasets, 2) inconsistent evaluation pipelines, 3) lacking workload diversity, and 4) lacking efficient comparison. Overall, there lacks an empirical study that puts TGNN models onto the same ground and compares them comprehensively. To this end, we propose Benchtemp, a general benchmark for evaluating TGNN models on various workloads. Benchtemp provides a set of benchmark datasets so that different TGNN models can be fairly compared. Further, Benchtemp engineers a standard pipeline that unifies the TGNN evaluation. With Benchtemp, we extensively compare the representative TGNN models on different tasks (e.g., link prediction and node classification) and settings (transductive and inductive), w.r.t. both effectiveness and efficiency metrics. We have made Benchtemp publicly available at https://github.com/qianghuangwhu/benchtemp and datasets at https://zenodo.org/record/8267846.

Heterogeneous graphs contain rich semantic information that can be exploited by heterogeneous graph neural networks (HGNNs). However, scaling HGNNs to large graphs is challenging due to the high computational cost. Existing non-parametric HGNNs use general subgraphs construction method and mean aggregator before training to reduce the complexity. Despite their success, they ignore two key characteristics of heterogeneous graphs, leading to low predictive performance. First, they adopt fixed local and global knowledge extractor for the feature aggregation and the semantic fusion. Besides, they bury the graph structure information of the higher-order meta-paths and fail to explore deeper graph structure information. In this paper, we address these two limitations and propose a new non-parametric HGNN framework called Heterogeneous Graph Attention Multi-Layer Perceptron (HGAMLP). Our framework employs the local multi-knowledge extractor to enhance the node representation, and leverages the de-redundancy mechanism to extract the pure graph structure information from higher-order meta-paths. Besides, it adopts a node-adaptive weight adjustment mechanism as an efficiency training model to fuse global knowledge and local knowledge. We evaluate our framework on ten commonly used heterogeneous graph datasets and show that it outperforms the state-of-the-art baselines in both accuracy and speed. Notably, our framework achieves the best performance on the large public heterogeneous graph dataset (i.e., Ogbn-mag) of Open Graph Benchmark11https://ogb.stanford.edu/docs/leader_nodeprop.

For graph self-supervised learning (GSSL), masked autoencoder (MAE) follows the generative paradigm and learns to reconstruct masked graph edges or node features while contrastive learning (CL) maximizes the similarity between augmented views of the same graph. Existing works utilize MAE and CL separately but we observe that the MAE and CL paradigms are complementary and propose the graph contrastive masked autoencoder (GCMAE) framework to unify them. Specifically, by focusing on local edges or node features, MAE cannot capture global information of the graph and is sensitive to particular edges and features. On the contrary, CL excels in extracting global information because it considers the relation between graphs. As such, we equip GCMAE with an MAE branch and a CL branch, and the two branches share a common encoder, which allows the MAE branch to exploit the global information extracted by the CL branch. To force GCMAE to capture global graph structures, we train it to reconstruct the entire adjacency matrix instead of only the masked edges as in existing works. Moreover, a discrimination loss is proposed for feature reconstruction, which improves the disparity between node embeddings rather than reducing the reconstruction error to tackle the feature smoothing problem of MAE. We evaluate GCMAE on four popular graph tasks (i.e., node classification, node clustering, link prediction, and graph classification) and compare it with 14 state-of-the-art baselines. The results show that GCMAE consistently provides good accuracy across these tasks, and the maximum accuracy improvement is up to 3.2% compared with the best-performing baseline.

Graph dataset condensation (GDC) reduces a dataset with many graphs into a smaller dataset with fewer graphs while maintaining model training accuracy. GDC saves the storage cost and hence accelerates training. Although several GDC methods have been proposed, they are all supervised and require massive labels for the graphs, while graph labels can be scarce in many practical scenarios. To fill this gap, we propose a self-supervised graph dataset condensation method called SGDC, which does not require label information. Our initial design starts with the classical bilevel optimization paradigm for dataset condensation and incorporates contrastive learning techniques. But such a solution yields poor accuracy due to the biased gradient estimation caused by data augmentation. To solve this problem, we introduce representation matching, which conducts training by aligning the representations produced by the condensed graphs with the target representations generated by a pre-trained SSL model. This design eliminates the need for data augmentation and avoids biased gradient. We further propose a graph attention kernel, which not only improves accuracy but also reduces running time when combined with self-supervised kernel ridge regression (KRR). To simplify SGDC and make it more robust, we adopt a adjacency matrix reusing approach, which reuses the topology of the original graphs for the condensed graphs instead of repeatedly learning topology during training. Our evaluations on seven graph datasets find that SGDC improves model accuracy by up to 9.7% compared with 5 state-of-the-art baselines, even if they use label information. Moreover, SGDC is significantly more efficient than the baselines.

One of the paradigms under which split learning (SL) is used is for the vertical federated learning (VFL) setting, where two or more parties build models over feature-partitioned data. However, to protect the private labels of one party, random noises are needed to perturb the backward derivatives (i.e., gradients w.r.t. forward activations), which incurs the privacy-utility tradeoff. In this work, we introduce PROJPERT,, a novel algorithm that searches for the optimal “perturbation knobs” for label protection in SL-based VFL. We formulate the problem of perturbation searching as how to minimize the impact on model quality given the desired privacy guarantee. Based on the problem, two solutions are introduced, where the first obtains the optimal perturbation via a simple but effective binary searching scheme, and the second heuristically approximates the optimality within a negligible error bound. Empirical results demonstrate that both our solutions are more effective in protecting the labels and achieve significantly better privacy-utility tradeoffs than state-of-the-art perturbation-based label protection methods. Furthermore, our heuristic solution is very efficient and incurs almost zero extra overhead in the overall running time, improving the usability in real-world applications.

Abstract Modern machine learning (ML) systems commonly use stochastic gradient descent (SGD) to train ML models. However, SGD relies on random data order to converge, which usually requires a full data shuffle. For in-DB ML systems and deep learning systems with large datasets stored onblock-addressable secondary storagesuch as HDD and SSD, this full data shuffle leads to low I/O performance—the data shuffling time can be even longer than the training itself, due to massive random data accesses. To balance the convergence rate of SGD (which favors data randomness) and its I/O performance (which favors sequential access), previous work has proposed several data shuffling strategies. In this paper, we first perform an empirical study on existing data shuffling strategies, showing that these strategies suffer from either low performance or low convergence rate. To solve this problem, we propose a simple but noveltwo-leveldata shuffling strategy named , which canavoida full data shuffle while maintainingcomparableconvergence rate of SGD as if a full shuffle were performed. We further theoretically analyze the convergence behavior of and empirically evaluate its efficacy in both in-DB ML and deep learning systems. For in-DB ML systems, we integrate into PostgreSQL by introducing three newphysicaloperators with optimizations. For deep learning systems, we extend single-process to multi-process for the parallel/distributed environment and integrate it into PyTorch. Our evaluation shows that can achieve comparable convergence rate with the full-shuffle-based SGD for both linear models and deep learning models. For in-DB ML with linear models, is 1.6 $$\times $$ × $$-$$ - 12.8 $$\times $$ × faster than two state-of-the-art systems, Apache MADlib and Bismarck, on both HDD and SSD. For deep learning models on ImageNet, is 1.5 $$\times $$ × faster than PyTorch with full data shuffle.

Graph convolutional networks (GCNs) have been successfully applied in many different real-world tasks. However, most of the existing methods are based on shallow GCN, because multiple layers involve long-distance neighborhood information but lead to the over-smoothing problem. Actually, a similar challenge exists in the depth limitation for primitive convolutional neural networks (CNNs). As the multi-layer architecture can increase the representation ability of GCN, we study and learn from the recent progress in CNN and propose Lasagne, a novel multi-layer GCN framework, empowered by node-aware layer aggregators and factorization-based layer interactions to overcome the over-smoothing problem and realize the full potentials of the GCN model. We analyze how the node locality affects the information propagation in GCN and propose a novel node aggregation mechanism in an adaptive manner. We further demystify Lasagne from a mutual information view and evaluate it on both real-world benchmark data sets and large-scale industrial production data sets. Lasagne shows strong empirical performance on the semi-supervised node classification task and outperforms the state-of-the-art methods without considering the node locality.

Detecting fraudulent transactions is an essential component to control risk in e-commerce marketplaces. Apart from rule-based and machine learning filters that are already deployed in production, we want to enable efficient real-time inference with graph neural networks (GNNs), which is useful to catch multihop risk propagation in a transaction graph. However, two challenges arise in the implementation of GNNs in production. First, future information in a dynamic graph should not be considered in message passing to predict the past. Second, the latency of graph query and GNN model inference is usually up to hundreds of milliseconds, which is costly for some critical online services. To tackle these challenges, we propose a Batch and Real-time Inception GrapH Topology (BRIGHT) framework to conduct an end-to-end GNN learning that allows efficient online real-time inference.

In this paper, we propose Lasagne, a novel multi-layer graph convolutional network (GCN) framework to over-come the over-smoothing problem and realize the full poten-tials of deep GCNs. We analyze how node localities affect the information propagation in GCN, propose an adaptive novel node aggregation mechanism and further demystify from a mutual information view. Evaluation results on both real-world benchmark data sets and large-scale industrial production data sets show Lasagne significantly outperforms the state-of- the-art methods without considering the node locality.

Network motif is a kind of frequently occurring subgraph that reflects local topology in graphs. Although network motif has been studied in graph analytics, e.g., social network and biological network, it is yet unclear whether network motif is useful for analyzing online transaction network that is generated in applications such as instant messaging and e-commerce. In this work, we analyze online transaction networks from the perspective of network motif. We define vertex features based on size-2 and size-3 motifs, and introduce motif-based centrality measurements. We further design motif-based vertex embedding that integrates weighted motif counts and centrality measurements. Afterward, we implement a distributed framework for motif detection in large-scale online transaction networks. To understand the effectiveness of motif for analyzing online transaction network, we study the statistical distribution of motifs in various kinds of graphs in Tencent and assess the benefit of motif-based embedding in a range of downstream graph analytical tasks. Empirical results show that our proposed method can efficiently find motifs in large-scale graphs, help interpretability, and benefit downstream tasks.

Stochastic gradient descent (SGD) is the cornerstone of modern ML systems. Despite its computational efficiency, SGD requires random data access that is inherently inefficient when implemented in systems that rely on block-addressable secondary storage such as HDD and SSD, e.g., in-DB ML systems and TensorFlow/PyTorch over large files. To address this impedance mismatch, various data shuffling strategies have been proposed to balance the convergence rate of SGD (which favors randomness) and its I/O performance (which favors sequential access).

Vertical federated learning (VFL) is an emerging paradigm that allows different parties (e.g., organizations or enterprises) to collaboratively build machine learning models with privacy protection. In the training phase, VFL only exchanges the intermediate statistics, i.e., forward activations and backward derivatives, across parties to compute model gradients. Nevertheless, due to its geo-distributed nature, VFL training usually suffers from the low WAN bandwidth. In this paper, we introduce CELU-VFL, a novel and efficient VFL training framework that exploits the local update technique to reduce the cross-party communication rounds. CELU-VFL caches the stale statistics and reuses them to estimate model gradients without exchanging the ad hoc statistics. Significant techniques are proposed to improve the convergence performance. First, to handle the stochastic variance problem, we propose a uniform sampling strategy to fairly choose the stale statistics for local updates. Second, to harness the errors brought by the stateness, we devise an instance weighting mechanism that measures the reliability of the estimated gradients. Theoretical analysis proves that CELU-VFL achieves a similar sub-linear convergence rate as vanilla VFL training but requires much fewer communication rounds. Empirical results on both public and real-world workloads validate that CELU-VFL can be up to six times faster than the existing works.

Wide models such as generalized linear models and factorization-based models have been extensively used in various predictive applications, e.g., recommendation, CTR prediction, and image recognition. Due to the memory bounded property of the models, the performance improvement on CPU is reaching the limitation. GPU is known to have many computation units and high memory bandwidth, and becomes a promising platform for training machine learning models. However, the GPU training for the wide models is far from optimal due to the sparsity and irregularity in wide models. The existing GPU-based wide models are even slower than the ones using CPU. The classical training schema of the wide models does not optimized for the GPU architecture, which suffers from large amount of random memory accesses and redundant read/write of intermediate values. In this paper, we propose an efficient GPU-training framework for the large-scale wide models, named cuWide. To fully benefit from the memory hierarchy of GPU, cuWide applies a new flow-based schema for training, which leverages the spatial and temporal locality of wide models to drastically reduce the amount of communication with GPU global memory. To do so, we adopt a bigraph computation model to efficiently realize the flow-based schema and exploit three flexible interfaces for programming. Further, we use the 2D partition of mini-batch (in sample and feature dimensions) with proposed graph abstraction to optimize GPU memory access for sparse data, and apply several spatial-temporal caching mechanisms (importance-based model caching and cross-stage accumulation caching mechanisms) to achieve a high performance kernel. To efficiently implement cuWide, we also propose several GPU-oriented optimizations, including feature-oriented data layout to enhance the data locality, replication mechanism to reduce update conflicts in shared memory, and multi-stream scheduling to overlap data transferring and kernel computing. We show that cuWide can be up to more than 20× faster than the state-of-the-art GPU solutions and multi-core CPU solutions.

In this paper, we propose an efficient GPU-training framework for the large-scale wide models, named cuWide. To fully benefit from the memory hierarchy of GPU, cuWide applies a new flow-based schema for training, which leverages the spatial and temporal locality of wide models to drastically reduce the amount of communication with GPU global memory. Comprehensive experiments show that cuWide can be up to more than 20× faster than the state-of-the-art GPU solutions and multi-core CPU solutions.

Black-box optimization (BBO) has a broad range of applications, including automatic machine learning, engineering, physics, and experimental design. However, it remains a challenge for users to apply BBO methods to their problems at hand with existing software packages, in terms of applicability, performance, and efficiency. In this paper, we build OpenBox, an open-source and general-purpose BBO service with improved usability. The modular design behind OpenBox also facilitates flexible abstraction and optimization of basic BBO components that are common in other existing systems. OpenBox is distributed, fault-tolerant, and scalable. To improve efficiency, OpenBox further utilizes "algorithm agnostic" parallelization and transfer learning. Our experimental results demonstrate the effectiveness and efficiency of OpenBox compared to existing systems.

With the ever-evolving concerns on privacy protection, vertical federated learning (FL), where participants own non-overlapping features for the same set of instances, is becoming a heated topic since it enables multiple enterprises to strengthen the machine learning models collaboratively with privacy guarantees. Nevertheless, to achieve privacy preservation, vertical FL algorithms involve complicated training routines and time-consuming cryptography operations, leading to slow training speed.

The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML). Several systems exist for training large-scale ML models on top of serverless infrastructures (e.g., AWS Lambda) but with inconclusive results in terms of their performance and relative advantage over "serverful" infrastructures (IaaS). In this paper we present a systematic, comparative study of distributed ML training over FaaS and IaaS. We present a design space covering design choices such as optimization algorithms and synchronization protocols, and implement a platform, LambdaML, that enables a fair comparison between FaaS and IaaS. We present experimental results using LambdaML, and further develop an analytic model to capture cost/performance tradeoffs that must be considered when opting for a serverless infrastructure. Our results indicate that ML training pays off in serverless only for models with efficient (i.e., reduced) communication and that quickly converge. In general, FaaS can be much faster but it is never significantly cheaper than IaaS.

All-reduce is the key communication primitive used in distributed data-parallel training due to the high performance in the homogeneous environment. However, All-reduce is sensitive to stragglers and communication delays as deep learning has been increasingly deployed on the heterogeneous environment like cloud. In this paper, we propose and analyze a novel variant of all-reduce, called partial-reduce, which provides high heterogeneity tolerance and performance by decomposing the synchronous all-reduce primitive into parallel-asynchronous partial-reduce operations. We provide theoretical guarantees, proving that partial-reduce converges to a stationary point at the similar sub-linear rate as distributed SGD. To enforce the convergence of the partial-reduce primitive, we further propose a dynamic staleness-aware distributed averaging algorithm and implement a novel group generation mechanism to prevent possible update isolation in heterogeneous environments. We build a prototype system in the real production cluster and validate its performance under different workloads. The experiments show that it is 1.21x-2x faster than other state-of-the-art baselines.

Recent years have witnessed a growing list of systems for distributed data-parallel training. Existing systems largely fit into two paradigms, i.e., parameter server and MPI-style collective operations. On the algorithmic side, researchers have proposed a wide range of techniques to lower the communication via "system relaxations": quantization, decentralization, and communication delay. However, most, if not all, existing systems only rely on standard synchronous and asynchronous stochastic gradient (SG) based optimization, therefore, cannot take advantage of all possible optimizations that the machine learning community has been developing recently. Given this emerging gap between the current landscapes of systems and theory, we build Bagua, a MPI-style communication library, providing a collection of primitives, that is both flexible and modular to support state-of-the-art system relaxation techniques of distributed training. Powered by this design, Bagua has a great ability to implement and extend various state-of-the-art distributed learning algorithms. In a production cluster with up to 16 machines (128 GPUs), Bagua can outperform PyTorch-DDP, Horovod and BytePS in the end-to-end training time by a significant margin (up to 2X) across a diverse range of tasks. Moreover, we conduct a rigorous tradeoff exploration showing that different algorithms and system relaxations achieve the best performance over different network conditions.

End-to-end AutoML has attracted intensive interests from both academia and industry, which automatically searches for ML pipelines in a space induced by feature engineering, algorithm/model selection, and hyper-parameter tuning. Existing AutoML systems, however, suffer from scalability issues when applying to application domains with large, high-dimensional search spaces. We present VOLCANOML, a scalable and extensible framework that facilitates systematic exploration of large AutoML search spaces. VOLCANOML introduces and implements basic building blocks that decompose a large search space into smaller ones, and allows users to utilize these building blocks to compose an execution plan for the AutoML problem at hand. VOLCANOML further supports a Volcano-style execution model - akin to the one supported by modern database systems - to execute the plan constructed. Our evaluation demonstrates that, not only does VOLCANOML raise the level of expressiveness for search space decomposition in AutoML, it also leads to actual findings of decomposition strategies that are significantly more efficient than the ones employed by state-of-the-art AutoML systems such as auto-sklearn.

Spark has extensively used in many applications of Tencent, due to its easy deployment, pipeline capability, and close integration with the Hadoop ecosystem. As the graph computing engine of Spark, GraphX is also widely deployed to process large-scale graph data in Tencent. However, when the size of the graph data is up to billion-scale, GraphX encounters serious performance degradation. Worse, Graphx cannot support the rising advancement of graph embedding (GE) and graph neural network (GNN) algorithms. To address these challenges, we develop a new graph processing system, called PSGraph, which uses Spark executor and PyTorch to perform calculation, and develops a distributed parameter server to store frequently accessed models. PSGraph can train extremely large-scale graph data in Tencent with the parameter server architecture, and enable the training of GE and GNN algorithms. Moreover, PSGraph still benefits from the advantages of Spark via staying inside the Spark ecosystem, and can directly replace GraphX without modification to the existing application framework. Our experiments show that PSGraph outperforms GraphX significantly.

Distributed machine learning (ML) has triggered tremendous research interest in recent years. Stochastic gradient descent (SGD) is one of the most popular algorithms for training ML models, and has been implemented in almost all distributed ML systems, such as Spark MLlib, Petuum, MXNet, and TensorFlow. However, current implementations often incur huge communication and memory overheads when it comes to large models. One important reason for this inefficiency is the row-oriented scheme (RowSGD) that existing systems use to partition the training data, which forces them to adopt a centralized model management strategy that leads to vast amount of data exchange over the network. We propose a novel, column-oriented scheme (ColumnSGD) that partitions training data by columns rather than by rows. As a result, ML model can be partitioned by columns as well, leading to a distributed configuration where individual data and model partitions can be collocated on the same machine. Following this locality property, we develop a simple yet powerful computation framework that significantly reduces communication overheads and memory footprints compared to RowSGD, for large-scale ML models such as generalized linear models (GLMs) and factorization machines (FMs). We implement ColumnSGD on top of Apache Spark, and study its performance both analytically and experimentally. Experimental results on both public and real-world datasets show that ColumnSGD is up to 930× faster than MLlib, 63× faster than Petuum, and 14× faster than MXNet.

The ensemble of deep neural networks has been shown, both theoretically and empirically, to improve generalization accuracy on the unseen test set. However, the high training cost hinders its efficiency since we need a sufficient number of base models and each one in the ensemble has to be separately trained. Lots of methods are proposed to tackle this problem, and most of them are based on the feature that a pre-trained network can transfer its knowledge to the next base model and then accelerate the training process. However, these methods suffer a severe problem that all of them transfer knowledge without selection and thus lead to low diversity. As the effect of ensemble learning is more pronounced if ensemble members are accurate and diverse, we propose a method named Efficient Diversity-Driven Ensemble (EDDE) to address both the diversity and the efficiency of an ensemble. To accelerate the training process, we propose a novel knowledge transfer method which can selectively transfer the previous generic knowledge. To enhance diversity, we first propose a new diversity measure, then use it to define a diversity-driven loss function for optimization. At last, we adopt a Boosting-based framework to combine the above operations, such a method can also further improve diversity. We evaluate EDDE on Computer Vision (CV) and Natural Language Processing (NLP) tasks. Compared with other well-known ensemble methods, EDDE can get highest ensemble accuracy with the lowest training cost, which means it is efficient in the ensemble of neural networks.

Graph Convolutional Network (GCN) is a widely used method for learning from graph-based data. However, it fails to use the unlabeled data to its full potential, thereby hindering its ability. Given some pseudo labels of the unlabeled data, the GCN can benefit from this extra supervision. Based on Knowledge Distillation and Ensemble Learning, lots of methods use a teacher-student architecture to make better use of the unlabeled data and then make a better prediction. However, these methods introduce unnecessary training costs and a high bias of student model if the teacher's predictions are unreliable. Besides, the final ensemble gains are limited due to limited diversity in the combined models. Therefore, we propose Reliable Data Distillation, a reliable data driven semi-supervised GCN training method. By defining the node reliability and edge reliability in a graph, we can make better use of high quality data and improve the graph representation learning. Furthermore, considering the data reliability and data importance, we propose a new ensemble learning method for GCN and a novel Self-Boosting SSL Framework to combine the above optimizations. Finally, our extensive evaluation of Reliable Data Distillation on real-world datasets shows that our approach outperforms the state-of-the-art methods on semi-supervised node classification tasks.

In Tencent Inc., more than 80% of the data are extracted and transformed using Spark. However, the commonly used machine learning systems are TensorFlow, XGBoost, and Angel, whereas Spark MLlib, an official Spark package for machine learning, is seldom used. One reason for this ignorance is that it is generally believed that Spark is slow when it comes to distributed machine learning. Users therefore have to undergo the painful procedure of moving data in and out of Spark. The question why Spark is slow, however, remains elusive. In this paper, we study the performance of MLlib with a focus on training generalized linear models using gradient descent. Based on a detailed examination, we identify two bottlenecks in MLlib, i.e., pattern of model update and pattern of communication. To address these two bottlenecks, we tweak the implementation of MLlib with two state-of-the-art and well-known techniques, model averaging and AllReduce. We show that, the new system that we call MLlib*, can significantly improve over MLlib and achieve similar or even better performance than other specialized distributed machine learning systems (such as Petuum and Angel), on both public and Tencent's workloads.

Most of the data is extracted and processed by Spark in Tencent Machine Learning Platform. However, seldom of them use Spark MLlib, an official machine learning (ML) library on top of Spark due to its inefficiency. In contrast, systems like parameter servers, XGBoost and TensorFlow are more used, which incur expensive cost of transferring data in and out of Spark ecosystem. In this paper, we identify the causes of inefficiency in Spark MLlib and solve the problem by building parameter servers on top of Spark. We propose PS2, a parameter server architecture that integrates Spark without hacking the core of Spark. With PS2, we leverage the power of Spark for data processing and ML training, and parameter servers for maintaining ML models. By carefully analyzing Tencent ML workloads, we figure out a widely existing computation pattern for ML models---element-wise operations among multiple high dimensional vectors. Based on this observation, we propose a new data abstraction, called Dimension Co-located Vector (DCV) for efficient model management in PS2. A DCV is a distributed vector that considers locality in parameter servers and enables efficient computation with multiple co-located distributed vectors. For ease-of-use, we also design a wide variety of advanced operators for operating DCVs. Finally, we carefully implement the PS2 system and evaluate it against existing systems on both public and Tencent workloads. Empirical results demonstrate that PS2 can outperform Spark MLlib by up to 55.6X and specialized ML systems like Petuum by up to 3.7X.

Gradient boosting decision tree (GBDT) is a widely-used machine learning algorithm in both data analytic competitions and real-world industrial applications. Further, driven by the rapid increase in data volume, efforts have been made to train GBDT in a distributed setting to support large-scale workloads. However, we find it surprising that the existing systems manage the training dataset in different ways, but none of them have studied the impact of data management. To that end, this paper aims to study the pros and cons of different data management methods regarding the performance of distributed GBDT. We first introduce a quadrant categorization of data management policies based on data partitioning and data storage. Then we conduct an in-depth systematic analysis and summarize the advantageous scenarios of the quadrants. Based on the analysis, we further propose a novel distributed GBDT system named Vero, which adopts the unexplored composition of vertical partitioning and row-store and suits for many large-scale cases. To validate our analysis empirically, we implement different quadrants in the same code base and compare them under extensive workloads, and finally compare Vero with other state-of-the-art systems over a wide range of datasets. Our theoretical and experimental results provide a guideline on choosing a proper data management policy for a given workload.

Gradient boosting decision tree (GBDT) is one of the most popular machine learning models widely used in both academia and industry. Although GBDT has been widely supported by existing systems such as XGBoost, LightGBM, and MLlib, one system bottleneck appears when the dimensionality of the data becomes high. As a result, when we tried to support our industrial partner on datasets of the dimension up to 330K, we observed suboptimal performance for all these aforementioned systems. In this paper, we ask "Can we build a scalable GBDT training system whose performance scales better with respect to dimensionality of the data?"

To address the challenge of explosive big data, distributed machine learning (ML) has drawn the interests of many researchers. Since many distributed ML algorithms trained by stochastic gradient descent (SGD) involve communicating gradients through the network, it is important to compress the transferred gradient. A category of low-precision algorithms can significantly reduce the size of gradients, at the expense of some precision loss. However, existing low-precision methods are not suitable for many cases where the gradients are sparse and nonuniformly distributed. In this paper, we study is there a compression method that can efficiently handle a sparse and nonuniform gradient consisting of key-value pairs?

Gradient boosting tree (GBT), a widely used machine learning algorithm, achieves state-of-the-art performance in academia, industry, and data analytics competitions. Although existing scalable systems which implement GBT, such as XGBoost and MLlib, perform well for datasets with medium-dimensional features, they can suffer performance degradation for many industrial applications where the trained datasets contain highdimensional features. The performance degradation derives from their inefficient mechanisms for model aggregation-either mapreduce or all-reduce. To address this high-dimensional problem, we propose a scalable execution plan using the parameter server architecture to facilitate the model aggregation. Further, we introduce a sparse-pull method and an efficient index structure to increase the processing speed. We implement a GBT system, namely TencentBoost, in the production cluster of Tencent Inc. The empirical results show that our system is 2-20× faster than existing platforms.

We study distributed machine learning in heterogeneous environments in this work. We first conduct a systematic study of existing systems running distributed stochastic gradient descent; we find that, although these systems work well in homogeneous environments, they can suffer performance degradation, sometimes up to 10x, in heterogeneous environments where stragglers are common because their synchronization protocols cannot fit a heterogeneous setting. Our first contribution is a heterogeneity-aware algorithm that uses a constant learning rate schedule for updates before adding them to the global parameter. This allows us to suppress stragglers' harm on robust convergence. As a further improvement, our second contribution is a more sophisticated learning rate schedule that takes into consideration the delayed information of each update. We theoretically prove the valid convergence of both approaches and implement a prototype system in the production cluster of our industrial partner Tencent Inc. We validate the performance of this prototype using a range of machine-learning workloads. Our prototype is 2-12x faster than other state-of-the-art systems, such as Spark, Petuum, and TensorFlow; and our proposed algorithm takes up to 6x fewer iterations to converge.

The exponential increase of online videos greatly enriches the life of users but also brings huge numbers of near-duplicate videos (NDVs) that seriously challenge the video websites. The video websites entail NDV-related applications such as detection of copyright violation, video monitoring, video re-ranking, and video recommendation. Since these applications adopt different features and different processing procedures due to diverse scenarios, constructing separate and special-purpose systems for them incurs considerable costs on design, implementation, and maintenance. In this article, we propose a general NDV system on Storm (GVoS)—a popular distributed real-time stream processing platform—to simultaneously support a wide variety of video applications. The generality of GVoS is achieved in two aspects. First, we extract the reusable components from various applications. Second, we conduct the communication between components via a mechanism called Stream Shared Message (SSM) that contains the video-related data. Furthermore, we present an algorithm to reduce the size of SSM in order to avoid the data explosion and decrease the network latency. The experimental results demonstrate that GVoS can achieve performance almost the same as the customized systems. Meanwhile, GVoS accomplishes remarkably higher systematic versatility and efficiently facilitates the development of various NDV-related applications.