Çağatay Demiralp

Robert Redmond, Nathan W. Weckwerth, Brian S. Xia et al.

We recently proposed a new cluster operating system stack, DBOS, centered on a DBMS. DBOS enables unique support for ML applications by encapsulating ML code within stored procedures, centralizing ancillary ML data, providing security built into the underlying DBMS, co-locating ML code and data, and tracking data and workflow provenance. Here we demonstrate a subset of these benefits around two ML applications. We first show that image classification and object detection models using GPUs can be served as DBOS stored procedures with performance competitive to existing systems. We then present a 1D CNN trained to detect anomalies in HTTP requests on DBOS-backed web services, achieving SOTA results. We use this model to develop an interactive anomaly detection system and evaluate it through qualitative user feedback, demonstrating its usefulness as a proof of concept for future work to develop learned real-time security services on top of DBOS.

Peter Baile Chen, Fabian Wenz, Yi Zhang et al.

Existing text-to-SQL benchmarks have largely been constructed from web tables with human-generated question-SQL pairs. LLMs typically show strong results on these benchmarks, leading to a belief that LLMs are effective at text-to-SQL tasks. However, how these results transfer to enterprise settings is unclear because tables in enterprise databases might differ substantially from web tables in structure and content. To contend with this problem, we introduce a new dataset BEAVER, the first enterprise text-to-SQL benchmark sourced from real private enterprise data warehouses. This dataset includes natural language queries and their correct SQL statements, which we collected from actual query logs. We then benchmark off-the-shelf LLMs on this dataset. LLMs perform poorly, even when augmented with standard prompt engineering and RAG techniques. We identify three main reasons for the poor performance: (1) schemas of enterprise tables are more complex than the schemas in public data, resulting in SQL-generation tasks intrinsically harder; (2) business-oriented questions are often more complex, requiring joins over multiple tables, aggregations, and nested queries; (3) public LLMs cannot train on private enterprise data warehouses that are not publicly accessible, and therefore it is difficult for the model to learn to solve (1) and (2). We believe BEAVER will facilitate future research in building text-to-SQL systems that perform better in enterprise settings.

Sajjadur Rahman, Peter Griggs, Çağatay Demiralp

With the increase in scale and availability of digital text generated on the web, enterprises such as online retailers and aggregators often use text analytics to mine and analyze the data to improve their services and products alike. Text data analysis is an iterative, non-linear process with diverse workflows spanning multiple stages, from data cleaning to visualization. Existing text analytics systems usually accommodate a subset of these stages and often fail to address challenges related to data heterogeneity, provenance, workflow reusability and reproducibility, and compatibility with established practices. Based on a set of design considerations we derive from these challenges, we propose Leam, a system that treats the text analysis process as a single continuum by combining advantages of computational notebooks, spreadsheets, and visualization tools. Leam features an interactive user interface for running text analysis workflows, a new data model for managing multiple atomic and composite data types, and an expressive algebra that captures diverse sets of operations representing various stages of text analysis and enables coordination among different components of the system, including data, code, and visualizations. We report our current progress in Leam development while demonstrating its usefulness with usage examples. Finally, we outline a number of enhancements to Leam and identify several research directions for developing an interactive visual text analysis system.

Sara Evensen, Chang Ge, Dongjin Choi et al.

Data programming is a programmatic weak supervision approach to efficiently curate large-scale labeled training data. Writing data programs (labeling functions) requires, however, both programming literacy and domain expertise. Many subject matter experts have neither programming proficiency nor time to effectively write data programs. Furthermore, regardless of one's expertise in coding or machine learning, transferring domain expertise into labeling functions by enumerating rules and thresholds is not only time consuming but also inherently difficult. Here we propose a new framework, data programming by demonstration (DPBD), to generate labeling rules using interactive demonstrations of users. DPBD aims to relieve the burden of writing labeling functions from users, enabling them to focus on higher-level semantics such as identifying relevant signals for labeling tasks. We operationalize our framework with Ruler, an interactive system that synthesizes labeling rules for document classification by using span-level annotations of users on document examples. We compare Ruler with conventional data programming through a user study conducted with 10 data scientists creating labeling functions for sentiment and spam classification tasks. We find that Ruler is easier to use and learn and offers higher overall satisfaction, while providing discriminative model performances comparable to ones achieved by conventional data programming.

Kevin Hu, Neil Gaikwad, Michiel Bakker et al.

Researchers currently rely on ad hoc datasets to train automated visualization tools and evaluate the effectiveness of visualization designs. These exemplars often lack the characteristics of real-world datasets, and their one-off nature makes it difficult to compare different techniques. In this paper, we present VizNet: a large-scale corpus of over 31 million datasets compiled from open data repositories and online visualization galleries. On average, these datasets comprise 17 records over 3 dimensions and across the corpus, we find 51% of the dimensions record categorical data, 44% quantitative, and only 5% temporal. VizNet provides the necessary common baseline for comparing visualization design techniques, and developing benchmark models and algorithms for automating visual analysis. To demonstrate VizNet's utility as a platform for conducting online crowdsourced experiments at scale, we replicate a prior study assessing the influence of user task and data distribution on visual encoding effectiveness, and extend it by considering an additional task: outlier detection. To contend with running such studies at scale, we demonstrate how a metric of perceptual effectiveness can be learned from experimental results, and show its predictive power across test datasets.

Marco Cavallo, Çağatay Demiralp

Dimensionality reduction is a common method for analyzing and visualizing high-dimensional data. However, reasoning dynamically about the results of a dimensionality reduction is difficult. Dimensionality-reduction algorithms use complex optimizations to reduce the number of dimensions of a dataset, but these new dimensions often lack a clear relation to the initial data dimensions, thus making them difficult to interpret. Here we propose a visual interaction framework to improve dimensionality-reduction based exploratory data analysis. We introduce two interaction techniques, forward projection and backward projection, for dynamically reasoning about dimensionally reduced data. We also contribute two visualization techniques, prolines and feasibility maps, to facilitate the effective use of the proposed interactions. We apply our framework to PCA and autoencoder-based dimensionality reductions. Through data-exploration examples, we demonstrate how our visual interactions can improve the use of dimensionality reduction in exploratory data analysis.

Selim Kalaycı, Çağatay Demiralp, Zeynep H. Gümüş

Precision oncology tests that profile tumors to identify clinically actionable targets have rapidly entered clinical practice. Effective visual presentation of the results of these tests is crucial in accurate clinical decision-making. In current practice, these results are typically delivered to oncologists as static prints, who then incorporate them into their clinical decision-making process. However, due to a lack of guidelines for standardization, different vendors use different report formats. There is very little known on the effectiveness of these report formats or the criteria necessary to improve them. In this study, we have aimed to identify both the tasks and the needs of oncologists from precision oncology report design and then to improve the designs based on these findings. To this end, we report results from multiple interviews and a survey study (n=32) conducted with practicing oncologists. Based on these results, we compiled a set of design criteria for precision oncology reports and developed a prototype report design using these criteria, along with feedback from oncologists.

Bahador Saket, Dominik Moritz, Halden Lin et al.

In this paper, we describe a research agenda for deriving design principles directly from data. We argue that it is time to go beyond manually curated and applied visualization design guidelines. We propose learning models of visualization design from data collected using graphical perception studies and build tools powered by the learned models. To achieve this vision, we need to 1) develop scalable methods for collecting training data, 2) collect different forms of training data, 3) advance interpretability of machine learning models, and 4) develop adaptive models that evolve as more data becomes available.

Marco Cavallo, Çağatay Demiralp

With the rapid commoditization of wearable sensors, detecting human movements from sensor datasets has become increasingly common over a wide range of applications. To detect activities, data scientists iteratively experiment with different classifiers before deciding which model to deploy. Effective reasoning about and comparison of alternative classifiers are crucial in successful model development. This is, however, inherently difficult in developing classifiers for sensor data, where the intricacy of long temporal sequences, high prediction frequency, and imprecise labeling make standard evaluation methods relatively ineffective and even misleading. We introduce Track Xplorer, an interactive visualization system to query, analyze, and compare the predictions of sensor-data classifiers. Track Xplorer enables users to interactively explore and compare the results of different classifiers, and assess their accuracy with respect to the ground-truth labels and video. Through integration with a version control system, Track Xplorer supports tracking of models and their parameters without additional workload on model developers. Track Xplorer also contributes an extensible algebra over track representations to filter, compose, and compare classification outputs, enabling users to reason effectively about classifier performance. We apply Track Xplorer in a collaborative project to develop classifiers to detect movements from multisensor data gathered from Parkinson's disease patients. We demonstrate how Track Xplorer helps identify early on possible systemic data errors, effectively track and compare the results of different classifiers, and reason about and pinpoint the causes of misclassifications.

Victor Dibia, Çağatay Demiralp

Rapidly creating effective visualizations using expressive grammars is challenging for users who have limited time and limited skills in statistics and data visualization. Even high-level, dedicated visualization tools often require users to manually select among data attributes, decide which transformations to apply, and specify mappings between visual encoding variables and raw or transformed attributes. In this paper we introduce Data2Vis, a neural translation model for automatically generating visualizations from given datasets. We formulate visualization generation as a sequence to sequence translation problem where data specifications are mapped to visualization specifications in a declarative language (Vega-Lite). To this end, we train a multilayered attention-based recurrent neural network (RNN) with long short-term memory (LSTM) units on a corpus of visualization specifications. Qualitative results show that our model learns the vocabulary and syntax for a valid visualization specification, appropriate transformations (count, bins, mean) and how to use common data selection patterns that occur within data visualizations. Data2Vis generates visualizations that are comparable to manually-created visualizations in a fraction of the time, with potential to learn more complex visualization strategies at scale.

Marco Cavallo, Çağatay Demiralp

Data clustering is a common unsupervised learning method frequently used in exploratory data analysis. However, identifying relevant structures in unlabeled, high-dimensional data is nontrivial, requiring iterative experimentation with clustering parameters as well as data features and instances. The number of possible clusterings for a typical dataset is vast, and navigating in this vast space is also challenging. The absence of ground-truth labels makes it impossible to define an optimal solution, thus requiring user judgment to establish what can be considered a satisfiable clustering result. Data scientists need adequate interactive tools to effectively explore and navigate the large clustering space so as to improve the effectiveness of exploratory clustering analysis. We introduce \textit{Clustrophile~2}, a new interactive tool for guided clustering analysis. \textit{Clustrophile~2} guides users in clustering-based exploratory analysis, adapts user feedback to improve user guidance, facilitates the interpretation of clusters, and helps quickly reason about differences between clusterings. To this end, \textit{Clustrophile~2} contributes a novel feature, the Clustering Tour, to help users choose clustering parameters and assess the quality of different clustering results in relation to current analysis goals and user expectations. We evaluate \textit{Clustrophile~2} through a user study with 12 data scientists, who used our tool to explore and interpret sub-cohorts in a dataset of Parkinson's disease patients. Results suggest that \textit{Clustrophile~2} improves the speed and effectiveness of exploratory clustering analysis for both experts and non-experts.

Marco Cavallo, Çağatay Demiralp

Detecting motor activities from sensor datasets is becoming increasingly common in a wide range of applications with the rapid commoditization of wearable sensors. To detect activities, data scientists iteratively experiment with different classifiers before deciding on a single model. Evaluating, comparing, and reasoning about prediction results of alternative classifiers is a crucial step in the process of iterative model development. However, standard aggregate performance metrics (such as accuracy score) and textual display of individual event sequences have limited granularity and scalability to effectively perform this critical step. To ameliorate these limitations, we introduce Track Xplorer, an interactive visualization system to query, analyze and compare the classification output of activity detection in multi-sensor data. Track Xplorer visualizes the results of different classifiers as well as the ground truth labels and the video of activities as temporally-aligned linear tracks. Through coordinated track visualizations, Track Xplorer enables users to interactively explore and compare the results of different classifiers, assess their accuracy with respect to the ground truth labels and video. Users can brush arbitrary regions of any classifier track, zoom in and out with ease, and playback the corresponding video segment to contextualize the performance of the classifier within the selected region. Track Xplorer also contributes an algebra over track representations to filter, compose, and compare classification outputs, enabling users to effectively reason about the performance of classifiers. We demonstrate how our tool helps data scientists debug misclassifications and improve the prediction performance in developing activity classifiers for real-world, multi-sensor data gathered from Parkinson's patients.

Çağatay Demiralp, Peter J. Haas, Srinivasan Parthasarathy et al.

Current tools for exploratory data analysis (EDA) require users to manually select data attributes, statistical computations and visual encodings. This can be daunting for large-scale, complex data. We introduce Foresight, a visualization recommender system that helps the user rapidly explore large high-dimensional datasets through "guideposts." A guidepost is a visualization corresponding to a pronounced instance of a statistical descriptor of the underlying data, such as a strong linear correlation between two attributes, high skewness or concentration about the mean of a single attribute, or a strong clustering of values. For each descriptor, Foresight initially presents visualizations of the "strongest" instances, based on an appropriate ranking metric. Given these initial guideposts, the user can then look at "nearby" guideposts by issuing "guidepost queries" containing constraints on metric type, metric strength, data attributes, and data values. Thus, the user can directly explore the network of guideposts, rather than the overwhelming space of data attributes and visual encodings. Foresight also provides for each descriptor a global visualization of ranking-metric values to both help orient the user and ensure a thorough exploration process. Foresight facilitates interactive exploration of large datasets using fast, approximate sketching to compute ranking metrics. We also contribute insights on EDA practices of data scientists, summarizing results from an interview study we conducted to inform the design of Foresight.

Bahador Saket, Alex Endert, Cagatay Demiralp

Visualizations of tabular data are widely used; understanding their effectiveness in different task and data contexts is fundamental to scaling their impact. However, little is known about how basic tabular data visualizations perform across varying data analysis tasks and data attribute types. In this paper, we report results from a crowdsourced experiment to evaluate the effectiveness of five visualization types --- Table, Line Chart, Bar Chart, Scatterplot, and Pie Chart --- across ten common data analysis tasks and three data attribute types using two real-world datasets. We found the effectiveness of these visualization types significantly varies across task and data attribute types, suggesting that visualization design would benefit from considering context dependent effectiveness. Based on our findings, we derive recommendations on which visualizations to choose based on different tasks. We finally train a decision tree on the data we collected to drive a recommender, showcasing how to effectively engineer experimental user data into practical visualization systems.