Modern deep learning models capture the semantics of complex data by transforming them into high-dimensional embedding vectors. Emerging applications, such as retrieval-augmented generation, use approximate nearest neighbor (ANN) search in the embedding vector space to find similar data. Existing vector databases provide indexes for efficient ANN searches, with graph-based indexes being the most popular due to their low latency and high recall in real-world high-dimensional datasets. However, these indexes are costly to build, suffer from significant contention under concurrent read-write workloads, and scale poorly to multiple servers. Our goal is to build a vector database that achieves high throughput and high recall under concurrent read-write workloads. To this end, we first propose an ANN index with an explicit two-stage design combining a fast filter stage with highly compressed vectors and a refine stage to ensure recall, and we devise a novel lightweight machine learning technique to fine-tune the index parameters. We introduce an early termination check to dynamically adapt the search process for each query. Next, we add support for writes while maintaining search performance by decoupling the management of the learned parameters. Finally, we design HAKES, a distributed vector database that serves the new index in a disaggregated architecture. We evaluate our index and system against 12 state-of-the-art indexes and three distributed vector databases, using high-dimensional embedding datasets generated by deep learning models. The experimental results show that our index outperforms index baselines in the high recall region and under concurrent read-write workloads. Furthermore, HAKES is scalable and achieves up to 16x higher throughputs than the baselines.

Fraud detection presents a challenging task characterized by ever-evolving fraud patterns and scarce labeled data. Existing methods predominantly rely on graph-based or sequence-based approaches. While graph-based approaches connect users through shared entities to capture structural information, they remain vulnerable to fraudsters who can disrupt or manipulate these connections. In contrast, sequence-based approaches analyze users' behavioral patterns, offering robustness against tampering but overlooking the interactions between similar users. Inspired by cohort analysis in retention and healthcare, this paper introduces VecAug, a novel cohort-augmented learning framework that addresses these challenges by enhancing the representation learning of target users with personalized cohort information. To this end, we first propose a vector burn-in technique for automatic cohort identification, which retrieves a task-specific cohort for each target user. Then, to fully exploit the cohort information, we introduce an attentive cohort aggregation technique for augmenting target user representations. To improve the robustness of such cohort augmentation, we also propose a novel label-aware cohort neighbor separation mechanism to distance negative cohort neighbors and calibrate the aggregated cohort information. By integrating this cohort information with target user representations, VecAug enhances the modeling capacity and generalization capabilities of the model to be augmented. Our framework is flexible and can be seamlessly integrated with existing fraud detection models. We deploy our framework on e-commerce platforms and evaluate it on three fraud detection datasets, and results show that VecAug improves the detection performance of base models by up to 2.48% in AUC and 22.5% in [email protected], outperforming state-of-the-art methods significantly.

Machine learning models are only as good as their training data. Simple models trained on well-chosen features extracted from the raw data often outperform complex models trained directly on the raw data. Data preparation pipelines, which clean and derive features from the data, are therefore important for machine learning applications. However, constructing such pipelines is a resource-intensive process that involves deep human expertise. Our goal is to design an efficient framework for automatically finding high-quality data preparation pipelines. The main challenge is how to explore a large search space of pipeline components with the objective of computing features that maximize the performance of the downstream models. Existing solutions are limited in terms of feature quality, which results in low accuracies of the downstream models, while incurring significant runtime overhead. We present CtxPipe, a novel framework that addresses the limitations of previous works by leveraging contextual information to improve the pipeline construction process. Specifically, it uses pre-trained embedding models to capture the data semantics, which are then used to guide the selection of pipeline components. We implement CtxPipe with deep reinforcement learning and evaluate it against state-of-the-art automated pipeline construction solutions. Our comprehensive experiments demonstrate that CtxPipe outperforms all of the baselines in both model performance and runtime cost.

The growing demand for advanced analytics beyond statistical aggregation calls for database systems that support effective model selection of deep neural networks (DNNs). However, existing model selection strategies are based on either training-based algorithms that deliver high-performing models at the expense of high computational cost, or training-free algorithms that enhance computational efficiency with reduced effectiveness. These strategies often disregard computational cost and response time Service-Level Objectives (SLOs), which are of concern to average or budget-conscious machine learning users. In addition, they lack a well-designed integration of the model selection algorithms with DBMSs, which hinders efficient in-database model selection. This paper presents TRAILS, a resource-efficient and SLO-aware in-database model selection system. To leverage the strengths of both training-free and training-based model selection, we first characterize nine state-of-the-art training-free model evaluation metrics and propose a more effective one named JacFlow, and then, restructure the conventional model selection procedure into two phases: filtering and refinement. A novel coordinator is also introduced to strike a balance between the high efficiency of train-free algorithms and the high effectiveness of training-based algorithms, ensuring high-performing model selection while adhering to target SLOs. Moreover, we incorporate the proposed algorithm into PostgreSQL to develop TRAILS, thereby both enhancing resource efficiency and reducing model selection latency. This integration establishes a foundation for declarative model definition and selection within DBMSs. Empirical results demonstrate that our TRAILS reduces model selection time and computational expenses considerably by up to 24.38x and 29.32x respectively compared to existing model selection systems.

Relational database management systems (RDBMS) are widely used for the storage of structured data. To derive insights beyond statistical aggregation, we typically have to extract specific subdatasets from the database using conventional database operations, and then apply deep neural networks (DNN) training and inference on these subdatasets in a separate analytics system. The process can be prohibitively expensive, especially when there are various subdatasets extracted for different analytical purposes. This calls for efficient in-database support of advanced analytical methods. In this paper, we introduce LEADS, a novel SQL-aware dynamic model slicing technique to customize models for specified SQL queries. LEADS improves the predictive modeling of structured data via the mixture of experts (MoE) and maintains efficiency by a SQL-aware gating network. At the core of LEADS is the construction of a general model with multiple expert sub-models trained over the database. The MoE scales up the modeling capacity, enhances effectiveness, and preserves efficiency by activating necessary experts via the SQL-aware gating network during inference. To support in-database analytics, we build an inference extension that integrates LEADS onto PostgreSQL. Our extensive experiments on real-world datasets demonstrate that LEADS consistently outperforms the baseline models, and the in-database inference extension delivers a considerable reduction in inference latency compared to traditional solutions.

Log messages provide a valuable source of runtime information for ensuring the safety and consistency of systems. Recently, many machine learning and deep learning methods have been proposed to automatically detect anomalous log messages, obviating the need for manual detection by experts. However, we find that in practice, the effectiveness of existing learning-based methods is severely affected by incomplete information and distribution shift. Specifically, each log message can actually be parsed into a fixed number of key information fields, while existing methods analyze log messages using only the log event information and ignore other useful information fields that can be critical to anomaly detection. Further, the distribution of real-world log messages changes continuously due to the dynamic nature of the runtime environment and thus, a detection model conventionally trained based on the unrealistic i.i.d. assumption may not provide the expected and consistent performance. In this paper, we present a robust and transferable anomaly detection framework RT-Log to address the above problems. To perform a comprehensive analysis of log messages, we introduce an adaptive relation modeling technique, which captures feature interactions among log information fields selectively and dynamically for effective and interpretable log representations. To establish its robustness and transferability, we propose a general environment generalization technique for learning the environment invariant representations that can generalize across different runtime environments. We evaluate the anomaly detection performance of RT-Log on large real-world datasets. Extensive experimental results demonstrate that RT-Log consistently outperforms state-of-the-art methods by a significant margin under different settings.

In line with the increasing machine learning model inference accuracy, deep learning (DL) models have been increasingly applied to structured data for a wide spectrum of real-world applications, including product recommendations, online advertisement, healthcare analytics and risk analysis. However, unlike unstructured data, structured data is high-dimensional and sparse and therefore engenders a large number of parameters in DL, making DL models more prone to overfitting. To alleviate the overfitting problem, various regularization methods have been designed to constrain the model parameters as a means to control the model complexity. Unfortunately, these methods are often restricted to regularizing the parameter values directly without considering the intrinsic correlations and dependencies between attribute fields of structured data which is however key to effective structured data modeling. In this paper, we re-examine DL for structured data from a new perspective of attribute interactions. In particular, we seek to explicitly model and regularize the pairwise relationships between attribute fields of structured data, in a field-adaptive manner, via a proposed attentive and interpretable framework called ATT-Reg. Specifically, in this framework, a set of attentive weight matrices are introduced to each attribute field for modeling obviously different relationships with its neighboring attribute fields. Further, we derive from the Bayesian viewpoint a novel Attentive Regularization method for imposing adaptive regularization strengths on different pairs of attribute fields, based on the informativeness of their relationship, which is calculated using both data-driven information and functional dependency (FD) knowledge. Such adaptive regularization facilitates each attribute field to learn discriminative and diversified representations for more effective predictive analytics. We also develop a feature attribution method for supporting more interpretable predictions We validate the effectiveness of our ATT-Reg on six real-world datasets. Extensive experimental results show that ATT-Reg achieves significant improvement over state-of-the-art graph models, attentive models as well as regularization methods and supports an excellent degree of interpretation.

Real-time analytics and decision-making require online anomaly detection (OAD) to handle drifts in data streams efficiently and effectively. Unfortunately, existing approaches are often constrained by their limited detection capacity and slow adaptation to evolving data streams, inhibiting their efficacy and efficiency in handling concept drift , which is a major challenge in evolving data streams. In this paper, we introduce METER, a novel dynamic concept adaptation framework that introduces a new paradigm for OAD. METER addresses concept drift by first training a base detection model on historical data to capture recurring central concepts , and then learning to dynamically adapt to new concepts in data streams upon detecting concept drift. Particularly, METER employs a novel dynamic concept adaptation technique that leverages a hypernetwork to dynamically generate the parameter shift of the base detection model, providing a more effective and efficient solution than conventional retraining or fine-tuning approaches. Further, METER incorporates a lightweight drift detection controller, underpinned by evidential deep learning, to support robust and interpretable concept drift detection. We conduct an extensive experimental evaluation, and the results show that METER significantly outperforms existing OAD approaches in various application scenarios.

In National University Hospital (NUH) in Singapore, we conduct healthcare analytics that analyzes heterogeneous electronic medical records (EMR) to support effective clinical decision-making on a daily basis. Existing work mainly focuses on multimodality for extracting complementary information from different modalities, and/or interpretability for providing interpretable prediction results. However, real-world healthcare analytics has presented another major challenge, i.e., the available modalities evolve or change intermittently. Addressing this challenge requires deployed models to be adaptive to such dynamic modality changes. To meet the aforementioned requirement, we develop a modular, multimodal and interpretable framework DyHealth to enable dynamic healthcare analytics in clinical practice. Specifically, different modalities are processed within their respective data modules that adhere to the interface defined by DyHealth. The extracted information from different modalities is integrated subsequently in our proposed Multimodal Fusion Module in DyHealth. In order to better handle modality changes at runtime, we further propose exponential increasing/decreasing mechanisms to support modality "hot-plug". We also devise a novel modality-based attention mechanism for providing fine-grained interpretation results on a per-input basis. We conduct a pilot evaluation of DyHealth on the patients' EMR data from NUH, in which DyHealth achieves superior performance and therefore, is promising to roll out for hospital-wide deployment. We also validate DyHealth in two public EMR datasets. Experimental results confirm the effectiveness, flexibility, and extensibility of DyHealth in supporting multimodal and interpretable healthcare analytics.

Relational databases are the de facto standard for storing and querying structured data, and extracting insights from structured data requires advanced analytics. Deep neural networks (DNNs) have achieved super-human prediction performance in particular data types, e.g., images. However, existing DNNs may not produce meaningful results when applied to structured data. The reason is that there are correlations and dependencies across combinations of attribute values in a table, and these do not follow simple additive patterns that can be easily mimicked by a DNN. The number of possible such cross features is combinatorial, making them computationally prohibitive to model. Furthermore, the deployment of learning models in real-world applications has also highlighted the need for interpretability, especially for high-stakes applications, which remains another issue of concern to DNNs. In this paper, we present ARM-Net, an adaptive relation modeling network tailored for structured data, and a lightweight framework ARMOR based on ARM-Net for relational data analytics. The key idea is to model feature interactions with cross features selectively and dynamically, by first transforming the input features into exponential space, and then determining the interaction order and interaction weights adaptively for each cross feature. We propose a novel sparse attention mechanism to dynamically generate the interaction weights given the input tuple, so that we can explicitly model cross features of arbitrary orders with noisy features filtered selectively. Then during model inference, ARM-Net can specify the cross features being used for each prediction for higher accuracy and better interpretability. Our extensive experiments on real-world datasets demonstrate that ARM-Net consistently outperforms existing models and provides more interpretable predictions for data-driven decision making.

Deep Learning and Machine Learning models have recently been shown to be effective in many real world applications. While these models achieve increasingly better predictive performance, their structures have also become much more complex. A common and difficult problem for complex models is overfitting. Regularization is used to penalize the complexity of the model in order to avoid overfitting. However, in most learning frameworks, regularization function is usually set with some hyper-parameters where the best setting is difficult to find. In this paper, we propose an adaptive regularization method, as part of a large end-to-end healthcare data analytics software stack, which effectively addresses the above difficulty. First, we propose a general adaptive regularization method based on Gaussian Mixture (GM) to learn the best regularization function according to the observed parameters. Second, we develop an effective update algorithm which integrates Expectation Maximization (EM) with Stochastic Gradient Descent (SGD). Third, we design a lazy update and sparse update algorithm to reduce the computational cost by 4x and 20x, respectively. The overall regularization framework is fast, adaptive, and easy-to-use. We validate the effectiveness of our regularization method through an extensive experimental study over 14 standard benchmark datasets and three kinds of deep learning/machine learning models. The results illustrate that our proposed adaptive regularization method achieves significant improvement over state-of-the-art regularization methods.

In high stakes applications such as healthcare and finance analytics, the interpretability of predictive models is required and necessary for domain practitioners to trust the predictions. Traditional machine learning models, e.g., logistic regression (LR), are easy to interpret in nature. However, many of these models aggregate time-series data without considering the temporal correlations and variations. Therefore, their performance cannot match up to recurrent neural network (RNN) based models, which are nonetheless difficult to interpret. In this paper, we propose a general framework TRACER to facilitate accurate and interpretable predictions, with a novel model TITV devised for healthcare analytics and other high stakes applications such as financial investment and risk management. Different from LR and other existing RNN-based models, TITV is designed to capture both the time-invariant and the time-variant feature importance using a feature-wise transformation subnetwork and a self-attention subnetwork, for the feature influence shared over the entire time series and the time-related importance respectively. Healthcare analytics is adopted as a driving use case, and we note that the proposed TRACER is also applicable to other domains, e.g., fintech. We evaluate the accuracy of TRACER extensively in two real-world hospital datasets, and our doctors/clinicians further validate the interpretability of TRACER in both the patient level and the feature level. Besides, TRACER is also validated in a critical financial application. The experimental results confirm that TRACER facilitates both accurate and interpretable analytics for high stakes applications.

With 5G on the verge of being adopted as the next mobile network, there is a need to analyze its impact on the landscape of computing and data management. In this paper, we analyze the impact of 5G on both traditional and emerging technologies and project our view on future research challenges and opportunities. With a predicted increase of 10-100× in bandwidth and 5-10x decrease in latency, 5G is expected to be the main enabler for smart cities, smart IoT and efficient healthcare, where machine learning is conducted at the edge. In this context, we investigate how 5G can help the development of federated learning. Network slicing, another key feature of 5G, allows running multiple isolated networks on the same physical infrastructure. However, security remains the main concern in the context of virtualization, multi-tenancy and high device density. Formal verification of 5G networks can be applied to detect security issues in massive virtualized environments. In summary, 5G will make the world even more densely and closely connected. What we have experienced in 4G connectivity will pale in comparison to the vast amounts of possibilities engendered by 5G.

Deep learning models have been used to support analytics beyond simple aggregation, where deeper and wider models have been shown to yield great results. These models consume a huge amount of memory and computational operations. However, most of the large-scale industrial applications are often computational budget constrained. In practice, the peak workload of inference service could be 10x higher than the average cases, with the presence of unpredictable extreme cases. Lots of computational resources could be wasted during off-peak hours and the system may crash when the workload exceeds system capacity. How to support deep learning services with dynamic workload cost-efficiently remains a challenging problem. In this paper, we address the challenge with a general and novel training scheme called model slicing , which enables deep learning models to provide predictions within the prescribed computational resource budget dynamically. Model slicing could be viewed as an elastic computation solution without requiring more computational resources. Succinctly, each layer in the model is divided into groups of contiguous block of basic components (i.e. neurons in dense layers and channels in convolutional layers), and then partially ordered relation is introduced to these groups by enforcing that groups participated in each forward pass always starts from the first group to the dynamically-determined rightmost group. Trained by dynamically indexing the rightmost group with a single parameter slice rate , the network is engendered to build up group-wise and residual representation. Then during inference, a sub-model with fewer groups can be readily deployed for efficiency whose computation is roughly quadratic to the width controlled by the slice rate. Extensive experiments show that models trained with model slicing can effectively support on-demand workload with elastic inference cost.

Deep Learning and Machine Learning models have recently been shown to be effective in many real world applications. While these models achieve increasingly better predictive performance, their structures have also become much more complex. A common and difficult problem for complex models is overfitting. Regularization is used to penalize the complexity of the model in order to avoid overfitting. However, in most learning frameworks, regularization function is usually set as some hyper parameters, and therefore the best setting is difficult to find. In this paper, we propose an adaptive regularization method, as part of a large end-to-end healthcare data analytics software stack, which effectively addresses the above difficulty. First, we propose a general adaptive regularization method based on Gaussian Mixture (GM) to learn the best regularization function according to the observed parameters. Second, we develop an effective update algorithm which integrates Expectation Maximization (EM) with Stochastic Gradient Descent (SGD). Third, we design a lazy update algorithm to reduce the computational cost by 4x. The overall regularization framework is fast, adaptive and easy-to-use. We validate the effectiveness of our regularization method through an extensive experimental study over 13 standard benchmark datasets and three kinds of deep learning/machine learning models. The results illustrate that our proposed adaptive regularization method achieves significant improvement over state-of-the-art regularization methods.