Although Multi-Vector Retrieval (MVR) has achieved the state of the art on many information retrieval (IR) tasks, its performance highly depends on how to decompose queries into smaller pieces, say phrases or tokens. However, optimizing query decomposition for MVR performance is not end-to-end differentiable. Even worse, jointly solving this problem and training the downstream retrieval-based systems, say RAG systems could be highly inefficient. To overcome these challenges, we propose Performance-Oriented Query Decomposer (POQD), a novel query decomposition framework for MVR. POQD leverages one LLM for query decomposition and searches the optimal prompt with an LLM-based optimizer. We further propose an end-to-end training algorithm to alternatively optimize the prompt for query decomposition and the downstream models. This algorithm can achieve superior MVR performance at a reasonable training cost as our theoretical analysis suggests. POQD can be integrated seamlessly into arbitrary retrieval-based systems such as Retrieval-Augmented Generation (RAG) systems. Extensive empirical studies on representative RAG-based QA tasks show that POQD outperforms existing query decomposition strategies in both retrieval performance and end-to-end QA accuracy. POQD is available at https://github.com/PKU-SDS-lab/POQD-ICML25.

With the proliferation of Large Language Models (LLMs) in Business Intelligence (BI), existing solutions face critical challenges in industrial deployments: functionality deficiencies from legacy systems failing to meet evolving LLM-era user demands, interaction limitations from single-round SQL generation paradigms inadequate for multi-round clarification, and cost for domain adaptation arising from cross-domain methods migration. We present SiriusBI, a practical LLM-powered BI system addressing the challenges of industrial deployments through three key innovations: (a) An end-to-end architecture integrating multi-module coordination to overcome functionality gaps in legacy systems; (b) A multi-round dialogue with querying mechanism, consisting of semantic completion, knowledge-guided clarification, and proactive querying processes, to resolve interaction constraints in SQL generation; (c) A data-conditioned SQL generation method selection strategy that supports both an efficient one-step Fine-Tuning approach and a two-step method leveraging Semantic Intermediate Representation for low-cost cross-domain applications. Experiments on both real-world datasets and public benchmarks demonstrate the effectiveness of SiriusBI. User studies further confirm that SiriusBI enhances both productivity and user experience. As an independent service on Tencent's data platform, SiriusBI is deployed across finance, advertising, and cloud sectors, serving dozens of enterprise clients. It achieves over 93% accuracy in SQL generation and reduces data analysts' query time from minutes to seconds in real-world applications.

How to effectively diagnose and mitigate database performance anomalies remains a significant concern for modern database systems. Manually identifying the root causes of the anomalies is a labor-intensive process and significantly relies on professional experience. Meanwhile, existing work on automatic database diagnosis mainly focuses on detecting anomalous performance metrics or system log. These solutions lack the power to pinpoint detailed issues such as bad queries or problematic operators, which are indispensable for most database troubleshooting processes. In this paper, we propose OpDiag, a diagnosis framework that attributes database performance anomalies to query operators. In this framework, we first construct models offline to represent the relationship between query operators, performance metrics, and anomalies. These models can capture query plan features and support ad-hoc queries and schemas. Then, through feature attribution on these models during online diagnosis, OpDiag can effectively identify critical anomalous metrics and further trace back to suspicious queries and operators. This can provide concrete guidance for subsequent steps in anomaly mitigation. We applied OpDiag to both synthetic benchmark and real industry cases from ZTE Corporation. Empirical studies prove that OpDiag can accurately localize anomalous queries and operators, thus reducing human efforts in diagnosing and mitigating database performance anomalies.

This paper researches on the power consumption of AIGC applications based on LLM with different parameter scales across different hardware platforms. Artificial Intelligence Generated Content (AIGC) represents a leading-edge application of AI technology, primarily driven by large language models (LLMs) and their associated technologies. The deployment of LLM typically relies on critical facilities with three layers, i.e., the hardware, model, and application layers. This empirical study aims to identify key factors in power consumption when a large model is serving in the inference stage, which will hint the insights for improving the energy efficiency of computational infrastructures. In the context of the "dual carbon" goals, i.e., carbon peaking and carbon neutrality, this study aims to find an effective way to reduce the energy cost of AIGC applications, thereby supporting sustainable AI development in industry.

Concept-based interpretability methods offer a lens into the internals of foundation models by decomposing their embeddings into high-level concepts. These concept representations are most useful when they are *compositional*, meaning that the individual concepts compose to explain the full sample. We show that existing unsupervised concept extraction methods find concepts which are not compositional. To automatically discover compositional concept representations, we identify two salient properties of such representations, and propose Compositional Concept Extraction (CCE) for finding concepts which obey these properties. We evaluate CCE on five different datasets over image and text data. Our evaluation shows that CCE finds more compositional concept representations than baselines and yields better accuracy on four downstream classification tasks.

Designing faithful yet accurate AI models is challenging, particularly in the field of individual treatment effect estimation (ITE). ITE prediction models deployed in critical settings such as healthcare should ideally be (i) accurate, and (ii) provide faithful explanations. However, current solutions are inadequate: state-of-the-art black-box models do not supply explanations, post-hoc explainers for black-box models lack faithfulness guarantees, and self-interpretable models greatly compromise accuracy. To address these issues, we propose DISCRET, a self-interpretable ITE framework that synthesizes faithful, rule-based explanations for each sample. A key insight behind DISCRET is that explanations can serve dually as database queries to identify similar subgroups of samples. We provide a novel RL algorithm to efficiently synthesize these explanations from a large search space. We evaluate DISCRET on diverse tasks involving tabular, image, and text data. DISCRET outperforms the best self-interpretable models and has accuracy comparable to the best black-box models while providing faithful explanations. DISCRET is available at https://github.com/wuyinjun-1993/DISCRET-ICML2024.

Finding errors in machine learning applications requires a thorough exploration of their behavior over data. Existing approaches used by practitioners are often ad-hoc and lack the abstractions needed to scale this process. We present TorchQL, a programming framework to evaluate and improve the correctness of machine learning applications. TorchQL allows users to write queries to specify and check integrity constraints over machine learning models and datasets. It seamlessly integrates relational algebra with functional programming to allow for highly expressive queries using only eight intuitive operators. We evaluate TorchQL on diverse use-cases including finding critical temporal inconsistencies in objects detected across video frames in autonomous driving, finding data imputation errors in time-series medical records, finding data labeling errors in real-world images, and evaluating biases and constraining outputs of language models. Our experiments show that TorchQL enables up to 13x faster query executions than baselines like Pandas and MongoDB, and up to 40% shorter queries than native Python. We also conduct a user study and find that TorchQL is natural enough for developers familiar with Python to specify complex integrity constraints.

Sample re-weighting strategies provide a promising mechanism to deal with imperfect training data in machine learning, such as noisily labeled or class-imbalanced data. One such strategy involves formulating a bi-level optimization problem called the meta re-weighting problem, whose goal is to optimize performance on a small set of perfect pivotal samples, called meta samples. Many approaches have been proposed to efficiently solve this problem. However, all of them assume that a perfect meta sample set is already provided while we observe that the selections of meta sample set is performance-critical. In this paper, we study how to learn to identify such a meta sample set from a large, imperfect training set, that is subsequently cleaned and used to optimize performance in the meta re-weighting setting. We propose a learning framework which reduces the meta samples selection problem to a weighted K-means clustering problem through rigorously theoretical analysis. We propose two clustering methods within our learning framework, Representation-based clustering method (RBC) and Gradient-based clustering method (GBC), for balancing performance and computational efficiency. Empirical studies demonstrate the performance advantage of our methods over various baseline methods

Machine learning models can make critical errors that are easily hidden within vast amounts of data. Such errors often run counter to rules based on human intuition. However, rules based on human knowledge are challenging to scale or to even formalize. We thereby seek to infer statistical rules from the data and quantify the extent to which a model has learned them. We propose a framework SQRL that integrates logic-based methods with statistical inference to derive these rules from a model’s training data without supervision. We further show how to adapt models at test time to reduce rule violations and produce more coherent predictions. SQRL generates up to 300K rules over datasets from vision, tabular, and language settings. We uncover up to 158K violations of those rules by state-of-the-art models for classification, object detection, and data imputation. Test-time adaptation reduces these violations by up to 68.7% with relative performance improvement up to 32%. SQRL is available at https://github.com/DebugML/sqrl.

Forecasting on sparse multivariate time series (MTS) aims to model the predictors of future values of time series given their incomplete past, which is important for many emerging applications. However, most existing methods process MTS’s individually, and do not leverage the dynamic distributions underlying the MTS’s, leading to sub-optimal results when the sparsity is high. To address this challenge, we propose a novel generative model, which tracks the transition of latent clusters, instead of isolated feature representations, to achieve robust modeling. It is characterized by a newly designed dynamic Gaussian mixture distribution, which captures the dynamics of clustering structures, and is used for emitting time series. The generative model is parameterized by neural networks. A structured inference network is also designed for enabling inductive analysis. A gating mechanism is further introduced to dynamically tune the Gaussian mixture distributions. Extensive experimental results on a variety of real-life datasets demonstrate the effectiveness of our method.

High-quality labels are expensive to obtain for many machine learning tasks, such as medical image classification tasks. Therefore, probabilistic (weak) labels produced by weak supervision tools are used to seed a process in which influential samples with weak labels are identified and cleaned by several human annotators to improve the model performance. To lower the overall cost and computational overhead of this process, we propose a solution called CHEF (CHEap and Fast label cleaning), which consists of the following three components. First, to reduce the cost of human annotators, we use INFL, which prioritizes the most influential training samples for cleaning and provides cleaned labels to save the cost of one human annotator. Second, to accelerate the sample selector phase and the model constructor phase, we use Increm-INFL to incrementally produce influential samples, and DeltaGrad-L to incrementally update the model. Third, we redesign the typical label cleaning pipeline so that human annotators iteratively clean smaller batch of samples rather than one big batch of samples. This yields better overall model performance and enables possible early termination when the expected model performance has been achieved. Extensive experiments show that our approach gives good model prediction performance while achieving significant speed-ups.

Non-volatile memory (NVM) is an emerging technology, which has the persistence characteristics of large capacity storage devices, while providing the low access latency and byte-addressablity of traditional DRAM memory. In this paper, we provide extensive performance evaluations on a recently released NVM device, Intel Optane DC Persistent Memory (PMem), under different configurations with several micro-benchmark tools. Further, we evaluate OLTP and OLAP database workloads with Microsoft SQL Server 2019 when using PMem as buffer pool or persistent storage. From the lessons learned we share some recommendations for future DBMS design with PMem, e.g. simple hardware or software changes are not enough for the best use of PMem in DBMSs.

Machine learning models are not static and may need to be retrained on slightly changed datasets, for instance, with the addition or deletion of a set of data points. This has many applications, including privacy, robustness, bias reduction, and uncertainty quantifcation. However, it is expensive to retrain models from scratch. To address this problem, we propose the DeltaGrad algorithm for rapid retraining machine learning models based on information cached during the training phase. We provide both theoretical and empirical support for the effectiveness of DeltaGrad, and show that it compares favorably to the state of the art.

The ubiquitous use of machine learning algorithms brings new challenges to traditional database problems such as incremental view update. Much effort is being put in better understanding and debugging machine learning models, as well as in identifying and repairing errors in training datasets. Our focus is on how to assist these activities when they have to retrain the machine learning model after removing problematic training samples in cleaning or selecting different subsets of training data for interpretability. This paper presents an efficient provenance-based approach, PrIU, and its optimized version, PrIU-opt, for incrementally updating model parameters without sacrificing prediction accuracy. We prove the correctness and convergence of the incrementally updated model parameters, and validate it experimentally. Experimental results show that up to two orders of magnitude speed-ups can be achieved by PrIU-opt compared to simply retraining the model from scratch, yet obtaining highly similar models.

As research products expand to include structured datasets, the challenge arises of how to automatically generate citations to the results of arbitrary queries against such datasets. Previous work explored this problem in the context of conjunctive queries and views using a Rewriting-Based Model (RBM). However, an increasing number of scientific queries are aggregate, e.g. statistical summaries of the underlying data, for which the RBM cannot be easily extended. In this paper, we show how a Provenance-Based Model (PBM) can be leveraged to 1) generate citations to conjunctive as well as aggregate queries and views; 2) associate citations with individual result tuples to enable arbitrary subsets of the result set to be cited ( fine-grained citations ); and 3) be optimized to return citations in acceptable time. Our implementation of PBM in ProvCite shows that it not only handles a larger class of queries and views than RBM, but can outperform it when restricted to conjunctive views in some cases.

An increasing amount of information is being published in structured databases and retrieved using queries, raising the question of how query results should be cited. Since there are a large number of possible queries over a database, one strategy is to specify citations to a small set of frequent queries - citation views - and use these to construct citations to other "general" queries. We present three approaches to implementing citation views and describe alternative policies for the joint, alternate and aggregated use of citation views. Extensive experiments using both synthetic and realistic citation views and queries show the trade-offs between the approaches in terms of the time to generate citations, as well as the size of the resulting citation. They also show that the choice of policy has a huge effect both on performance and size, leading to useful guidelines for what policies to use and how to specify citation views.

An increasing amount of information is being collected in structured, evolving, curated databases, driving the question of how information extracted from such datasets via queries should be cited. While several databases say how data should be cited for web-page views of the database, they leave it to users to manually construct the citations. Furthermore, they do not say how data extracted by queries other than web-page views -- general queries -- should be cited. This demo shows how citations can be specified for a small set of views of the database, and used to automatically generate citations for general queries against the database.

Efficient retrieval of traffic archival data is a must-have technique to detect network attacks, such as APT(advanced persistent threat) attack. In order to take insight from Internet traffic, the bitmap index is increasingly used for efficiently querying over large datasets. However, a raw bitmap index leads to high space consumption and overhead on loading indexes. Various bitmap index compression algorithms are proposed to save storage while improving query efficiency. This paper proposes a new bitmap index compression algorithm called BAH (Byte Aligned Hybrid compression coding). An acceleration algorithm using SIMD is designed to increase the efficiency of AND operation over multiple compressed bitmaps. In all, BAH has a better compression ratio and faster intersection querying speed compared with several previous works such as WAH, PLWAH, COMPAX, Roaring etc. The theoretical analysis shows that the space required by BAH is no larger than 1.6 times the information entropy of the bitmap with density larger than 0.2%. In the experiments, BAH saves about 65% space and 60% space compared with WAH on two datasets. The experiments also demonstrate the query efficiency of BAH with the application in Internet Traffic and Web pages.

Bitmap indexing is flexible to conduct boolean operations in data retrieval. Besides, the query processing based on bitmap indexing is also very fast. Therefore it has been widely used in various big data analytics platforms, such as Druid and Spark etc. However, bitmap index can consume a large amount of memory, which leads to the invention of different kinds of bitmap index compression algorithms without sacrificing temporal performance. In practice, we are often discommoded by choosing a proper algorithm when handling specific problems. Besides, after devising a new algorithm that may outperform existing ones, it is essential to evaluate its performance in theory. Without appropriate theoretical analysis, the deficit of a new algorithm can only be spotted until final experimental results are drawn, thus wasting much time and effort. In this paper, we propose a general analytical model to analyze both the spatial and temporal performance for bitmap index compression algorithms, which can be applied to analyze all kinds of algorithms derived from WAH (word-aligned hybrid). In this model, two types of distributed bitmaps, uniformly distributed bitmaps and clustered bitmaps, are used separately. In order to illustrate this model, several bitmap index compression algorithms are analyzed and compared with each other. Algorithms herein are COMBAT (COMbining Binary And Ternary encoding), SECOMPAX (Scope Extended COMPAX) and CONCISE (Compressed 'n' Composable Integer Set), which are all derived from WAH. Evaluation results by MATLAB simulation about these algorithms are also presented. This paper paves the way for further researches on the performance evaluation of various bitmap index compression algorithms in the future.