Search

In this work, we propose Table-LLM-Specialist, or Table-Specialist for short, as a new self-trained fine-tuning paradigm specifically designed for table tasks. Our insight is that for each table task, there often exist two dual versions of the same task, one generative and one classification in nature. Leveraging their duality, we propose a Generator-Validator paradigm, to iteratively generate-then-validate training data from language-models, to fine-tune stronger \sys models that can specialize in a given task, without requiring manually-labeled data. Our extensive evaluations suggest that our Table-Specialist has (1) \textit{strong performance} on diverse table tasks over vanilla language-models -- for example, Table-Specialist fine-tuned on GPT-3.5 not only outperforms vanilla GPT-3.5, but can often match or surpass GPT-4 level quality, (2) \textit{lower cost} to deploy, because when Table-Specialist fine-tuned on GPT-3.5 achieve GPT-4 level quality, it becomes possible to deploy smaller models with lower latency and inference cost, with comparable quality, and (3) \textit{better generalizability} when evaluated across multiple benchmarks, since \sys is fine-tuned on a broad range of training data systematically generated from diverse real tables. Our code and data will be available at https://github.com/microsoft/Table-LLM-Specialist.

Spreadsheets are widely recognized as the most popular end-user programming tools, which blend the power of formula-based computation, with an intuitive table-based interface. Today, spreadsheets are used by billions of users to manipulate tables, most of whom are neither database experts nor professional programmers. Despite the success of spreadsheets, authoring complex formulas remains challenging, as non-technical users need to look up and understand non-trivial formula syntax. To address this pain point, we leverage the observation that there is often an abundance of similar-looking spreadsheets in the same organization, which not only have similar data, but also share similar computation logic encoded as formulas. We develop an Auto-Formula system that can accurately predict formulas that users want to author in a target spreadsheet cell, by learning and adapting formulas that already exist in similar spreadsheets, using contrastive-learning techniques inspired by "similar-face recognition" from compute vision. Extensive evaluations on over 2K test formulas extracted from real enterprise spreadsheets show the effectiveness of Auto-Formula over alternatives. Our benchmark data is available at https://github.com/microsoft/Auto-Formula to facilitate future research., Comment: full version of a paper to appear in SIGMOD 2024

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.

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 analytics 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/Power-BI/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., Comment: full version of a paper accepted to VLDB 2023

Business Intelligence (BI) is crucial in modern enterprises and billion-dollar business. Traditionally, technical experts like database administrators would manually prepare BI-models (e.g., in star or snowflake schemas) that join tables in data warehouses, before less-technical business users can run analytics using end-user dashboarding tools. However, the popularity of self-service BI (e.g., Tableau and Power-BI) in recent years creates a strong demand for less technical end-users to build BI-models themselves. We develop an Auto-BI system that can accurately predict BI models given a set of input tables, using a principled graph-based optimization problem we propose called \textit{k-Min-Cost-Arborescence} (k-MCA), which holistically considers both local join prediction and global schema-graph structures, leveraging a graph-theoretical structure called \textit{arborescence}. While we prove k-MCA is intractable and inapproximate in general, we develop novel algorithms that can solve k-MCA optimally, which is shown to be efficient in practice with sub-second latency and can scale to the largest BI-models we encounter (with close to 100 tables). Auto-BI is rigorously evaluated on a unique dataset with over 100K real BI models we harvested, as well as on 4 popular TPC benchmarks. It is shown to be both efficient and accurate, achieving over 0.9 F1-score on both real and synthetic benchmarks., Comment: full version of a paper accepted to VLDB 2023

Data pipelines are widely employed in modern enterprises to power a variety of Machine-Learning (ML) and Business-Intelligence (BI) applications. Crucially, these pipelines are \emph{recurring} (e.g., daily or hourly) in production settings to keep data updated so that ML models can be re-trained regularly, and BI dashboards refreshed frequently. However, data quality (DQ) issues can often creep into recurring pipelines because of upstream schema and data drift over time. As modern enterprises operate thousands of recurring pipelines, today data engineers have to spend substantial efforts to \emph{manually} monitor and resolve DQ issues, as part of their DataOps and MLOps practices. Given the high human cost of managing large-scale pipeline operations, it is imperative that we can \emph{automate} as much as possible. In this work, we propose Auto-Validate-by-History (AVH) that can automatically detect DQ issues in recurring pipelines, leveraging rich statistics from historical executions. We formalize this as an optimization problem, and develop constant-factor approximation algorithms with provable precision guarantees. Extensive evaluations using 2000 production data pipelines at Microsoft demonstrate the effectiveness and efficiency of AVH., Comment: full version of a paper accepted to KDD 2023

As data lakes become increasingly popular in large enterprises today, there is a growing need to tag or classify data assets (e.g., files and databases) in data lakes with additional metadata (e.g., semantic column-types), as the inferred metadata can enable a range of downstream applications like data governance (e.g., GDPR compliance), and dataset search. Given the sheer size of today's enterprise data lakes with petabytes of data and millions of data assets, it is imperative that data assets can be ``auto-tagged'', using lightweight inference algorithms and minimal user input. In this work, we develop Auto-Tag, a corpus-driven approach that automates data-tagging of \textit{custom} data types in enterprise data lakes. Using Auto-Tag, users only need to provide \textit{one} example column to demonstrate the desired data-type to tag. Leveraging an index structure built offline using a lightweight scan of the data lake, which is analogous to pre-training in machine learning, Auto-Tag can infer suitable data patterns to best ``describe'' the underlying ``domain'' of the given column at an interactive speed, which can then be used to tag additional data of the same ``type'' in data lakes. The Auto-Tag approach can adapt to custom data-types, and is shown to be both accurate and efficient. Part of Auto-Tag ships as a ``custom-classification'' feature in a cloud-based data governance and catalog solution \textit{Azure Purview}.

Data analysis often involves comparing subsets of data across many dimensions for finding unusual trends and patterns. While the comparison between subsets of data can be expressed using SQL, they tend to be complex to write, and suffer from poor performance over large and high-dimensional datasets. In this paper, we propose a new logical operator COMPARE for relational databases that concisely captures the enumeration and comparison between subsets of data and greatly simplifies the expressing of a large class of comparative queries. We extend the database engine with optimization techniques that exploit the semantics of COMPARE to significantly improve the performance of such queries. We have implemented these extensions inside Microsoft SQL Server, a commercial DBMS engine. Our extensive evaluation on synthetic and real-world datasets shows that COMPARE results in a significant speedup over existing approaches, including physical plans generated by today's database systems, user-defined function (UDF), as well as middleware solutions that compare subsets outside the databases., Comment: Extended Version of VLDB 2021 paper

Recent work has made significant progress in helping users to automate single data preparation steps, such as string-transformations and table-manipulation operators (e.g., Join, GroupBy, Pivot, etc.). We in this work propose to automate multiple such steps end-to-end, by synthesizing complex data pipelines with both string transformations and table-manipulation operators. We propose a novel "by-target" paradigm that allows users to easily specify the desired pipeline, which is a significant departure from the traditional by-example paradigm. Using by-target, users would provide input tables (e.g., csv or json files), and point us to a "target table" (e.g., an existing database table or BI dashboard) to demonstrate how the output from the desired pipeline would schematically "look like". While the problem is seemingly underspecified, our unique insight is that implicit table constraints such as FDs and keys can be exploited to significantly constrain the space to make the problem tractable. We develop an Auto-Pipeline system that learns to synthesize pipelines using reinforcement learning and search. Experiments on large numbers of real pipelines crawled from GitHub suggest that Auto-Pipeline can successfully synthesize 60-70% of these complex pipelines with up to 10 steps., Comment: This is a full version of a conference paper to appear at VLDB 2021

Fuzzy similarity join is an important database operator widely used in practice. So far the research community has focused exclusively on optimizing fuzzy join \textit{scalability}. However, practitioners today also struggle to optimize fuzzy-join \textit{quality}, 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 \textsc{Auto-FuzzyJoin} framework that can infer suitable fuzzy-join programs on given input tables, without requiring explicit human input such as labeled training data. Using \textsc{Auto-FuzzyJoin}, users only need to provide two input tables $L$ and $R$, and a desired precision target $\tau$ (say 0.9). \textsc{Auto-FuzzyJoin} leverages the fact that one of the input is a reference table to automatically program fuzzy-joins that meet the precision target $\tau$ 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 \textsc{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., Comment: full version of a paper in SIGMOD 2021

Bitvector filtering is an important query processing technique that can significantly reduce the cost of execution, especially for complex decision support queries with multiple joins. Despite its wide application, however, its implication to query optimization is not well understood. In this work, we study how bitvector filters impact query optimization. We show that incorporating bitvector filters into query optimization straightforwardly can increase the plan space complexity by an exponential factor in the number of relations in the query. We analyze the plans with bitvector filters for star and snowflake queries in the plan space of right deep trees without cross products. Surprisingly, with some simplifying assumptions, we prove that, the plan of the minimal cost with bitvector filters can be found from a linear number of plans in the number of relations in the query. This greatly reduces the plan space complexity for such queries from exponential to linear. Motivated by our analysis, we propose an algorithm that accounts for the impact of bitvector filters in query optimization. Our algorithm optimizes the join order for an arbitrary decision support query by choosing from a linear number of candidate plans in the number of relations in the query. We implement our algorithm in Microsoft SQL Server as a transformation rule. Our evaluation on both industry standard benchmarks and customer workload shows that, compared with the original Microsoft SQL Server, our technique reduces the total CPU execution time by 22%-64% for the workloads, with up to two orders of magnitude reduction in CPU execution time for individual queries., Comment: This technical report is an extended version of the ACM SIGMOD 2020 paper Bitvector-aware Query Optimization for Decision Support Queries

The web contains a vast corpus of HTML tables. They can be used to provide direct answers to many web queries. We focus on answering two classes of queries with those tables: those seeking lists of entities (e.g., `cities in california') and those seeking superlative entities (e.g., `largest city in california'). The main challenge is to achieve high precision with significant coverage. Existing approaches train machine learning models to select the answer from the candidates; they rely on textual match features between the query and the content of the table along with features capturing table quality/importance. These features alone are inadequate for achieving the above goals. Our main insight is that we can improve precision by (i) first extracting intent (structured information) from the query for the above query classes and (ii) then performing structure-aware matching (instead of just textual matching) between the extracted intent and the candidates to select the answer. We model (i) as a sequence tagging task. We leverage state-of-the-art deep neural network models with word embeddings. The model requires large scale training data which is expensive to obtain via manual labeling; we therefore develop a novel method to automatically generate the training data. For (ii), we develop novel features to compute structure-aware match and train a machine learning model. Our experiments on real-life web search queries show that (i) our intent extractor for list and superlative intent queries has significantly higher precision and coverage compared with baseline approaches and (ii) our table answer selector significantly outperforms the state-of-the-art baseline approach. This technology has been used in production by Microsoft's Bing search engine since 2016.

A Kavli foundation sponsored workshop on the theme \emph{Petabytes to Science} was held 12$^{th}$ to 14$^{th}$ of February 2019 in Las Vegas. The aim of the this workshop was to discuss important trends and technologies which may support astronomy. We also tackled how to better shape the workforce for the new trends and how we should approach education and public outreach. This document was coauthored during the workshop and edited in the weeks after. It comprises the discussions and highlights many recommendations which came out of the workshop. We shall distill parts of this document and formulate potential white papers for the decadal survey., Comment: 70 pages 2 figures - this contains a few fixes

A machine learning configuration refers to a combination of preprocessor, learner, and hyperparameters. Given a set of configurations and a large dataset randomly split into training and testing set, we study how to efficiently select the best configuration with approximately the highest testing accuracy when trained from the training set. To guarantee small accuracy loss, we develop a solution using confidence interval (CI)-based progressive sampling and pruning strategy. Compared to using full data to find the exact best configuration, our solution achieves more than two orders of magnitude speedup, while the returned top configuration has identical or close test accuracy., Comment: Full version of an AAAI 2019 conference paper

In visual exploration and analysis of data, determining how to select and transform the data for visualization is a challenge for data-unfamiliar or inexperienced users. Our main hypothesis is that for many data sets and common analysis tasks, there are relatively few "data slices" that result in effective visualizations. By focusing human users on appropriate and suitably transformed parts of the underlying data sets, these data slices can help the users carry their task to correct completion. To verify this hypothesis, we develop a framework that permits us to capture exemplary data slices for a user task, and to explore and parse visual-exploration sequences into a format that makes them distinct and easy to compare. We develop a recommendation system, DataSlicer, that matches a "currently viewed" data slice with the most promising "next effective" data slices for the given exploration task. We report the results of controlled experiments with an implementation of the DataSlicer system, using four common analytical task types. The experiments demonstrate statistically significant improvements in accuracy and exploration speed versus users without access to our system.

We aim to provide table answers to keyword queries against knowledge bases. For queries referring to multiple entities, like "Washington cities population" and "Mel Gibson movies", it is better to represent each relevant answer as a table which aggregates a set of entities or entity-joins within the same table scheme or pattern. In this paper, we study how to find highly relevant patterns in a knowledge base for user-given keyword queries to compose table answers. A knowledge base can be modeled as a directed graph called knowledge graph, where nodes represent entities in the knowledge base and edges represent the relationships among them. Each node/edge is labeled with type and text. A pattern is an aggregation of subtrees which contain all keywords in the texts and have the same structure and types on node/edges. We propose efficient algorithms to find patterns that are relevant to the query for a class of scoring functions. We show the hardness of the problem in theory, and propose path-based indexes that are affordable in memory. Two query-processing algorithms are proposed: one is fast in practice for small queries (with small patterns as answers) by utilizing the indexes; and the other one is better in theory, with running time linear in the sizes of indexes and answers, which can handle large queries better. We also conduct extensive experimental study to compare our approaches with a naive adaption of known techniques., Comment: VLDB 2014

The ability to estimate resource consumption of SQL queries is crucial for a number of tasks in a database system such as admission control, query scheduling and costing during query optimization. Recent work has explored the use of statistical techniques for resource estimation in place of the manually constructed cost models used in query optimization. Such techniques, which require as training data examples of resource usage in queries, offer the promise of superior estimation accuracy since they can account for factors such as hardware characteristics of the system or bias in cardinality estimates. However, the proposed approaches lack robustness in that they do not generalize well to queries that are different from the training examples, resulting in significant estimation errors. Our approach aims to address this problem by combining knowledge of database query processing with statistical models. We model resource-usage at the level of individual operators, with different models and features for each operator type, and explicitly model the asymptotic behavior of each operator. This results in significantly better estimation accuracy and the ability to estimate resource usage of arbitrary plans, even when they are very different from the training instances. We validate our approach using various large scale real-life and benchmark workloads on Microsoft SQL Server., Comment: VLDB2012

The need for accurate SQL progress estimation in the context of decision support administration has led to a number of techniques proposed for this task. Unfortunately, no single one of these progress estimators behaves robustly across the variety of SQL queries encountered in practice, meaning that each technique performs poorly for a significant fraction of queries. This paper proposes a novel estimator selection framework that uses a statistical model to characterize the sets of conditions under which certain estimators outperform others, leading to a significant increase in estimation robustness. The generality of this framework also enables us to add a number of novel "special purpose" estimators which increase accuracy further. Most importantly, the resulting model generalizes well to queries very different from the ones used to train it. We validate our findings using a large number of industrial real-life and benchmark workloads., Comment: VLDB2012

Data analysis applications typically aggregate data across many dimensions looking for anomalies or unusual patterns. The SQL aggregate functions and the GROUP BY operator produce zero-dimensional or one-dimensional aggregates. Applications need the N-dimensional generalization of these operators. This paper defines that operator, called the data cube or simply cube. The cube operator generalizes the histogram, cross-tabulation, roll-up, drill-down, and sub-total constructs found in most report writers. The novelty is that cubes are relations. Consequently, the cube operator can be imbedded in more complex non-procedural data analysis programs. The cube operator treats each of the N aggregation attributes as a dimension of N-space. The aggregate of a particular set of attribute values is a point in this space. The set of points forms an N-dimensional cube. Super-aggregates are computed by aggregating the N-cube to lower dimensional spaces. This paper (1) explains the cube and roll-up operators, (2) shows how they fit in SQL, (3) explains how users can define new aggregate functions for cubes, and (4) discusses efficient techniques to compute the cube. Many of these features are being added to the SQL Standard.