Large Language Models (LLMs) have democratized database access through Text-to-SQL, but moving from prototypes to production remains difficult. Real deployments must handle strict SQL dialects, massive schemas, and evolving user preferences, while supervised fine-tuning is costly and rigid and agentic test-time scaling is expensive. We present Tahoe, a system that treats prompt optimization as a dynamic data management problem. Tahoe uses an error-driven hint learning pipeline across Development and Deployment to consolidate debugging traces into a structured Hint Bank. Compiler feedback is distilled into reusable Syntax Hints for dialect-specific rules, while execution and user feedback are converted into Semantic Hints for schema- and user-specific logic. Tahoe further introduces a Strategy Layer that models conflicting user intents as competing strategies under shared natural-language triggers, with recency signals and post-learning attribution statistics that summarize empirical success, harm, inertness, and support. At inference time, Tahoe retrieves relevant hints and guides the LLM through Logic Planning followed by SQL Synthesis. We implement and evaluate the development-phase workflow, leaving deployment-time human-feedback updates for future work. On Spider 2.0-Snow, Tahoe substantially improves Text-to-SQL without updating model parameters. On 113 supervised Spider 2.0-Snow-0212 examples using GPT-5.5, Tahoe raises pass rate from 61.95 percent to 79.42 percent and pass-at-4 from 72.57 percent to 87.61 percent, achieves 100 percent Snowflake syntax pass rate, and reduces average compiler-feedback critic rounds from 2.79 to 0.12 per sampled candidate. The same Hint Bank also transfers to weaker backbones, including a 19.7 percentage-point pass-rate gain on Doubao-2.0-lite.

Relational tables, where each row corresponds to an entity and each column corresponds to an attribute, have been the standard for tables in relational databases. However, such a standard cannot be taken for granted when dealing with tables "in the wild". Our survey of real spreadsheet-tables and web-tables shows that over 30% of such tables do not conform to the relational standard, for which complex table-restructuring transformations are needed before these tables can be queried easily using SQL-based tools. Unfortunately, the required transformations are non-trivial to program, which has become a substantial pain point for technical and non-technical users alike, as evidenced by large numbers of forum questions in places like StackOverflow and Excel/Tableau forums. We develop an Auto-Tables system that can automatically synthesize pipelines with multi-step transformations (in Python or other languages), to transform non-relational tables into standard relational forms for downstream analytics, obviating the need for users to manually program transformations. We compile an extensive benchmark for this new task, by collecting 244 real test cases from user spreadsheets and online forums. Our evaluation suggests that Auto-Tables can successfully synthesize transformations for over 70% of test cases at interactive speeds, without requiring any input from users, making this an effective tool for both technical and non-technical users to prepare data for analytics.

Relational tables, where each row corresponds to an entity and each column corresponds to an attribute, have been the standard for tables in relational databases. However, such a standard cannot be taken for granted when dealing with tables "in the wild". Our survey of real spreadsheettables and web-tables shows that over 30% of such tables do not conform to the relational standard, for which complex table-restructuring transformations are needed before these tables can be queried easily using SQL-based tools. Unfortunately, the required transformations are non-trivial to program, which has become a substantial pain point for technical and non-technical users alike, as evidenced by large numbers of forum questions in places like StackOverflow and Excel/Tableau forums.

Language models, such as GPT-3.5 and ChatGPT, demonstrate remarkable abilities to follow diverse human instructions and perform a wide range of tasks. However, when probing language models using a range of basic table-understanding tasks, we observe that today's language models are still sub-optimal in many table-related tasks, likely because they are pre-trained predominantly on \emph{one-dimensional} natural-language texts, whereas relational tables are \emph{two-dimensional} objects. In this work, we propose a new "\emph{table-tuning}" paradigm, where we continue to train/fine-tune language models like GPT-3.5 and ChatGPT, using diverse table-tasks synthesized from real tables as training data, with the goal of enhancing language models' ability to understand tables and perform table tasks. We show that our resulting Table-GPT models demonstrate (1) better \emph{table-understanding} capabilities, by consistently outperforming the vanilla GPT-3.5 and ChatGPT, on a wide-range of table tasks, including holdout unseen tasks, and (2) strong \emph{generalizability}, in its ability to respond to diverse human instructions to perform new table-tasks, in a manner similar to GPT-3.5 and ChatGPT.

Process mining has shown great impact in improving business Key Performance Indicators (KPIs), which are typically measured as aggregations over case-level outcomes. A commonly encountered key question in achieving such impact is understanding the underlying reasons for why a certain outcome appears in some cases (e.g., why certain cases take long to finish). We use the term drivers to refer to explanations for case-level outcomes. We hypothesize that how process is run, in other words, process traces, directly influences case-level outcomes, and hence KPIs. In this paper, we propose a new method to automatically and efficiently discover process-based drivers that are effective, significant and interpretable. We formally define the problem of driver discovery as a constrained optimization problem. Given that the problem is NP-hard, we develop efficient greedy algorithms to solve the problem. We evaluate our method on real-world datasets to demonstrate the effectiveness and efficiency of our approach.

We report our experience in running three editions (2020, 2021, 2022) of the SIGMOD programming contest, a well-known event for students to engage in solving exciting data management problems. During this period we had the opportunity of introducing participants to the entity resolution task, which is of paramount importance in the data integration community. We aim at sharing the executive decisions, made by the people coauthoring this report, and the lessons learned.

Data preprocessing is a crucial step in the machine learning process that transforms raw data into a more usable format for downstream ML models. However, it can be costly and time-consuming, often requiring the expertise of domain experts. Existing automated machine learning (AutoML) frameworks claim to automate data preprocessing. However, they often use a restricted search space of data preprocessing pipelines which limits the potential performance gains, and they are often too slow as they require training the ML model multiple times. In this paper, we propose DiffPrep, a method that can automatically and efficiently search for a data preprocessing pipeline for a given tabular dataset and a differentiable ML model such that the performance of the ML model is maximized. We formalize the problem of data preprocessing pipeline search as a bi-level optimization problem. To solve this problem efficiently, we transform and relax the discrete, non-differential search space into a continuous and differentiable one, which allows us to perform the pipeline search using gradient descent with training the ML model only once. Our experiments show that DiffPrep achieves the best test accuracy on 15 out of the 18 real-world datasets evaluated and improves the model's test accuracy by up to 6.6 percentage points.

Most real-world datasets contain label noise, which can negatively affect downstream ML models trained on them. To deal with this problem, one can clean the mislabeled data before training, which is not only time-consuming and expensive but also requires domain expertise. Another approach is to use a noise-robust ML training algorithm. However, existing methods have some prerequisites that may not be practical in many applications (e.g., they are tied to specific downstream model architecture or they are applicable to specific noise distributions). In this paper, we propose a model-agnostic approach for learning with noisy labels of arbitrary distributions. In particular, our approach can work with any gradient descent optimization based machine learning model and deal with any label noise distribution. We achieve them by proposing two theoretically grounded noise-robust loss functions (for different noise distributions), and we are able to automatically decide which loss function to use based on a novel noise setting detection module. We directly learn the required hyper-parameters in the loss functions via meta-learning technique to minimize the loss on a given small clean validation set, and propose several strategies to improve the efficiency of training. Experiments on multiple datasets with both real-world and injected label noise show that our method performs better than state-of-the-art approaches.

Entity matching (EM) refers to the problem of identifying tuple pairs in one or more relations that refer to the same real world entities. Supervised machine learning (ML) approaches, and deep learning based approaches in particular, typically achieve state-of-the-art matching results. However, these approaches require many labeled examples, in the form of matching and non-matching pairs, which are expensive and time-consuming to label. In this paper, we introduce Panda, a weakly supervised system specifically designed for EM. Panda uses the same labeling function abstraction as Snorkel, where labeling functions (LF) are user-provided programs that can generate large amounts of (somewhat noisy) labels quickly and cheaply, which can then be combined via a labeling model to generate accurate final predictions. To support users developing LFs for EM, Panda provides an integrated development environment (IDE) that lives in a modern browser architecture. Panda's IDE facilitates the development, debugging, and life-cycle management of LFs in the context of EM tasks, similar to how IDEs such as Visual Studio or Eclipse excel in general-purpose programming. Panda's IDE includes many novel features purpose-built for EM, such as smart data sampling, a builtin library of EM utility functions, automatically generated LFs, visual debugging of LFs, and finally, an EM-specific labeling model. We show in this demo that Panda IDE can greatly accelerate the development of high-quality EM solutions using weak supervision.

Fuzzy similarity join is an important database operator widely used in practice. So far the research community has focused exclusively on optimizing fuzzy joinscalability. However, practitioners today also struggle to optimize fuzzy-joinquality, because they face a daunting space of parameters (e.g., distance-functions, distance-thresholds, tokenization-options, etc.), and often have to resort to a manual trial-and-error approach to program these parameters in order to optimize fuzzy-join quality. This key challenge of automatically generating high-quality fuzzy-join programs has received surprisingly little attention thus far. In this work, we study the problem of "auto-program'' fuzzy-joins. Leveraging a geometric interpretation of distance-functions, we develop an unsupervised Auto-FuzzyJoin framework that can infer suitable fuzzy-join programs on given input tables, without requiring explicit human input such as labelled training data. Using Auto-FuzzyJoin, users only need to provide two input tables L and R, and a desired precision target τ (say 0.9). Auto-FuzzyJoin leverages the fact that one of the input is a reference table to automatically program fuzzy-joins that meet the precision target τ in expectation, while maximizing fuzzy-join recall (defined as the number of correctly joined records). Experiments on both existing benchmarks and a new benchmark with 50 fuzzy-join tasks created from Wikipedia data suggest that the proposed Auto-FuzzyJoin significantly outperforms existing unsupervised approaches, and is surprisingly competitive even against supervised approaches (e.g., Magellan and DeepMatcher) when 50% of ground-truth labels are used as training data. We have released our code and benchmark on GitHub\footnote\urlhttps://github.com/chu-data-lab/AutomaticFuzzyJoin to facilitate future research.

Data quality affects machine learning (ML) model performances, and data scientists spend considerable amount of time on data cleaning before model training. However, to date, there does not exist a rigorous study on how exactly cleaning affects ML - ML community usually focuses on developing ML algorithms that are robust to some particular noise types of certain distributions, while database (DB) community has been mostly studying the problem of data cleaning alone without considering how data is consumed by downstream ML analytics.We propose a CleanML study that systematically investigates the impact of data cleaning on ML classification tasks. The open-source and extensible CleanML study currently includes 14 real-world datasets with real errors, five common error types, seven different ML models, and multiple cleaning algorithms for each error type (including both commonly used algorithms in practice as well as state-of-the-art solutions in academic literature). We control the randomness in ML experiments using statistical hypothesis testing, and we also control false discovery rate in our experiments using the Benjamini-Yekutieli (BY) procedure. We analyze the results in a systematic way to derive many interesting and nontrivial observations. We also put forward multiple research directions for researchers.

Fuzzy similarity join is an important database operator widely used in practice. So far the research community has focused exclusively on optimizing fuzzy joinscalability. However, practitioners today also struggle to optimize fuzzy-joinquality, because they face a daunting space of parameters (e.g., distance-functions, distance-thresholds, tokenization-options, etc.), and often have to resort to a manual trial-and-error approach to program these parameters in order to optimize fuzzy-join quality. This key challenge of automatically generating high-quality fuzzy-join programs has received surprisingly little attention thus far. In this work, we study the problem of "auto-program'' fuzzy-joins. Leveraging a geometric interpretation of distance-functions, we develop an unsupervised Auto-FuzzyJoin framework that can infer suitable fuzzy-join programs on given input tables, without requiring explicit human input such as labelled training data. Using Auto-FuzzyJoin, users only need to provide two input tables L and R, and a desired precision target τ (say 0.9). Auto-FuzzyJoin leverages the fact that one of the input is a reference table to automatically program fuzzy-joins that meet the precision target τ in expectation, while maximizing fuzzy-join recall (defined as the number of correctly joined records). Experiments on both existing benchmarks and a new benchmark with 50 fuzzy-join tasks created from Wikipedia data suggest that the proposed Auto-FuzzyJoin significantly outperforms existing unsupervised approaches, and is surprisingly competitive even against supervised approaches (e.g., Magellan and DeepMatcher) when 50% of ground-truth labels are used as training data. We have released our code and benchmark on GitHub\footnote\urlhttps://github.com/chu-data-lab/AutomaticFuzzyJoin to facilitate future research.

Fuzzy join is an important database operator that is widely used in practice. So far the research community has focused exclusively on the important problem of efficient fuzzy join algorithms. Today, practitioners also struggle with a daunting space of parameters to optimize join quality (e.g., distance-function, distancethresholds, tokenization-options, etc.), where a manual trial-anderror approach has become a major pain-point. This important parameter-tuning problem has thus far received surprisingly little attention from the research community. Motivated by the observation, we study the problem of “autotune” suitable fuzzy-join parameters on given input tables, without requiring explicit human input such as labelled training data. We develop an Auto-FuzzyJoin framework that, when given two tables L and R, and a user-specified target precision τ (say 0.9), can automatically configure fuzzy-join parameters meeting the precision target τ in expectation, while maximizing fuzzy-join recall (defined as the number of correctly joined records). We study the hardness of this problem, and develop effective algorithms that are extensively tested on both existing benchmarks as well as a new benchmark of 50 fuzzy-join tasks we created from Wikipedia data. Our evaluation suggests that the proposed Auto-FuzzyJoin significantly outperforms existing unsupervised approaches (e.g., Microsoft Excel and ZeroER). It is surprisingly competitive even against supervised approaches (e.g., Magellan and DeepMatcher) when 50% of ground-truth joins are used as training data. We will release our code and benchmark to facilitate future research. ACM Reference Format: Peng Li, Xiang Cheng, Xu Chu, Yeye He, and Surajit Chaudhuri. 2021. AutoFuzzyJoin: Auto-Tune Fuzzy JoinsWithout Labeled Examples. In Proceedings ∗Both authors contributed equally to this research. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. SIGMOD ’21, June 20–25, 2021, Xi’an, Shaanxi, China © 2021 Association for Computing Machinery. ACM ISBN 978-x-xxxx-xxxx-x/YY/MM. . . $15.00 https://doi.org/10.1145/nnnnnnn.nnnnnnn of 2021 International Conference on Management of Data (SIGMOD ’21).ACM, New York, NY, USA, 16 pages. https://doi.org/10.1145/nnnnnnn.nnnnnnn

Entity matching (EM) refers to the problem of identifying tuple pairs in one or more relations that refer to the same real world entities. Supervised machine learning (ML) approaches, and deep learning based approaches in particular, typically achieve state-of-the-art matching results. However, these approaches require many labeled examples, in the form of matching and non-matching pairs, which are expensive and time-consuming to label. In this paper, we introduce Panda, a weakly supervised system specifically designed for EM. Panda uses the same labeling function abstraction as Snorkel, where labeling functions (LF) are user-provided programs that can generate large amounts of (somewhat noisy) labels quickly and cheaply, which can then be combined via a labeling model to generate accurate final predictions. To support users developing LFs for EM, Panda provides an integrated development environment (IDE) that lives in a modern browser architecture. Panda's IDE facilitates the development, debugging, and life-cycle management of LFs in the context of EM tasks, similar to how IDEs such as Visual Studio or Eclipse excel in general-purpose programming. Panda's IDE includes many novel features purpose-built for EM, such as smart data sampling, a builtin library of EM utility functions, automatically generated LFs, visual debugging of LFs, and finally, an EM-specific labeling model. We show in this demo that Panda IDE can greatly accelerate the development of high-quality EM solutions using weak supervision.

Machine learning (ML) applications have been thriving recently, largely attributed to the increasing availability of data. However, inconsistency and incomplete information are ubiquitous in real-world datasets, and their impact on ML applications remains elusive. In this paper, we present a formal study of this impact by extending the notion of Certain Answers for Codd tables , which has been explored by the database research community for decades, into the field of machine learning. Specifically, we focus on classification problems and propose the notion of "Certain Predictions" (CP) --- a test data example can be certainly predicted (CP'ed) if all possible classifiers trained on top of all possible worlds induced by the incompleteness of data would yield the same prediction. We study two fundamental CP queries: (Q1) checking query that determines whether a data example can be CP'ed; and (Q2) counting query that computes the number of classifiers that support a particular prediction (i.e., label). Given that general solutions to CP queries are, not surprisingly, hard without assumption over the type of classifier, we further present a case study in the context of nearest neighbor (NN) classifiers, where efficient solutions to CP queries can be developed --- we show that it is possible to answer both queries in linear or polynomial time over exponentially many possible worlds. We demonstrate one example use case of CP in the important application of "data cleaning for machine learning (DC for ML)." We show that our proposed CPClean approach built based on CP can often significantly outperform existing techniques, particularly on datasets with systematic missing values. For example, on 5 datasets with systematic missingness, CPClean (with early termination) closes 100% gap on average by cleaning 36% of dirty data on average, while the best automatic cleaning approach BoostClean can only close 14% gap on average.

Machine learning (ML) applications have been thriving recently, largely attributed to the increasing availability of data. However, inconsistency and incomplete information are ubiquitous in real-world datasets, and their impact on ML applications remains elusive. In this paper, we present a formal study of this impact by extending the notion of Certain Answers for Codd tables , which has been explored by the database research community for decades, into the field of machine learning. Specifically, we focus on classification problems and propose the notion of "Certain Predictions" (CP) --- a test data example can be certainly predicted (CP'ed) if all possible classifiers trained on top of all possible worlds induced by the incompleteness of data would yield the same prediction. We study two fundamental CP queries: (Q1) checking query that determines whether a data example can be CP'ed; and (Q2) counting query that computes the number of classifiers that support a particular prediction (i.e., label). Given that general solutions to CP queries are, not surprisingly, hard without assumption over the type of classifier, we further present a case study in the context of nearest neighbor (NN) classifiers, where efficient solutions to CP queries can be developed --- we show that it is possible to answer both queries in linear or polynomial time over exponentially many possible worlds. We demonstrate one example use case of CP in the important application of "data cleaning for machine learning (DC for ML)." We show that our proposed CPClean approach built based on CP can often significantly outperform existing techniques, particularly on datasets with systematic missing values. For example, on 5 datasets with systematic missingness, CPClean (with early termination) closes 100% gap on average by cleaning 36% of dirty data on average, while the best automatic cleaning approach BoostClean can only close 14% gap on average.

It is widely recognized that the data quality affects machine learning (ML) model performances, and data scientists spend considerable amount of time on data cleaning before model training. However, to date, there does not exist a rigorous study on how exactly does cleaning affect ML --- ML community usually focuses on the effects of specific types of noises of certain distributions (e.g., mislabels) on certain ML models, while database (DB) community has been mostly studying the problem of data cleaning alone without considering how data is consumed by downstream analytics. We propose the CleanML benchmark that systematically investigates the impact of data cleaning on downstream ML models. The CleanML benchmark currently includes 13 real-world datasets with real errors, five common error types, and seven different ML models. To ensure that our findings are statistically significant, CleanML carefully controls the randomness in ML experiments using statistical hypothesis testing, and also uses the Benjamini-Yekutieli (BY) procedure to control potential false discoveries due to many hypotheses in the benchmark. We obtain many interesting and non-trivial insights, and identify multiple open research directions. We also release the benchmark and hope to invite future studies on the important problems of joint data cleaning and ML.