Arvind Satyanarayan

Angie Boggust, Harini Suresh, Hendrik Strobelt et al. · mit

Saliency methods are a common class of machine learning interpretability techniques that calculate how important each input feature is to a model's output. We find that, with the rapid pace of development, users struggle to stay informed of the strengths and limitations of new methods and, thus, choose methods for unprincipled reasons (e.g., popularity). Moreover, despite a corresponding rise in evaluation metrics, existing approaches assume universal desiderata for saliency methods (e.g., faithfulness) that do not account for diverse user needs. In response, we introduce saliency cards: structured documentation of how saliency methods operate and their performance across a battery of evaluative metrics. Through a review of 25 saliency method papers and 33 method evaluations, we identify 10 attributes that users should account for when choosing a method. We group these attributes into three categories that span the process of computing and interpreting saliency: methodology, or how the saliency is calculated; sensitivity, or the relationship between the saliency and the underlying model and data; and, perceptibility, or how an end user ultimately interprets the result. By collating this information, saliency cards allow users to more holistically assess and compare the implications of different methods. Through nine semi-structured interviews with users from various backgrounds, including researchers, radiologists, and computational biologists, we find that saliency cards provide a detailed vocabulary for discussing individual methods and allow for a more systematic selection of task-appropriate methods. Moreover, with saliency cards, we are able to analyze the research landscape in a more structured fashion to identify opportunities for new methods and evaluation metrics for unmet user needs.

Benny J. Tang, Angie Boggust, Arvind Satyanarayan · mit

Captions that describe or explain charts help improve recall and comprehension of the depicted data and provide a more accessible medium for people with visual disabilities. However, current approaches for automatically generating such captions struggle to articulate the perceptual or cognitive features that are the hallmark of charts (e.g., complex trends and patterns). In response, we introduce VisText: a dataset of 12,441 pairs of charts and captions that describe the charts' construction, report key statistics, and identify perceptual and cognitive phenomena. In VisText, a chart is available as three representations: a rasterized image, a backing data table, and a scene graph -- a hierarchical representation of a chart's visual elements akin to a web page's Document Object Model (DOM). To evaluate the impact of VisText, we fine-tune state-of-the-art language models on our chart captioning task and apply prefix-tuning to produce captions that vary the semantic content they convey. Our models generate coherent, semantically rich captions and perform on par with state-of-the-art chart captioning models across machine translation and text generation metrics. Through qualitative analysis, we identify six broad categories of errors that our models make that can inform future work.

Angie Boggust, Hyemin Bang, Hendrik Strobelt et al. · mit

While interpretability methods identify a model's learned concepts, they overlook the relationships between concepts that make up its abstractions and inform its ability to generalize to new data. To assess whether models' have learned human-aligned abstractions, we introduce abstraction alignment, a methodology to compare model behavior against formal human knowledge. Abstraction alignment externalizes domain-specific human knowledge as an abstraction graph, a set of pertinent concepts spanning levels of abstraction. Using the abstraction graph as a ground truth, abstraction alignment measures the alignment of a model's behavior by determining how much of its uncertainty is accounted for by the human abstractions. By aggregating abstraction alignment across entire datasets, users can test alignment hypotheses, such as which human concepts the model has learned and where misalignments recur. In evaluations with experts, abstraction alignment differentiates seemingly similar errors, improves the verbosity of existing model-quality metrics, and uncovers improvements to current human abstractions.

Zoe De Simone, Angie Boggust, Fredo Durand et al. · mit

Text-to-image (T2I) systems enable rapid generation of high-fidelity imagery but are misaligned with how visual ideas develop. T2I systems generate outputs that make implicit visual decisions on behalf of the user, often introduce fine-grained details that can anchor users prematurely and limit their ability to keep options open early on, and cause unintended changes during editing that are difficult to correct and reduce users' sense of control. To address these concerns, we present Creo, a multi-stage T2I system that scaffolds image generation by progressing from rough sketches to high-resolution outputs, exposing intermediary abstractions where users can make incremental changes. Sketch-like abstractions invite user editing and allow users to keep design options open when ideas are still forming due to their provisional nature. Each stage in Creo can be modified with manual changes and AI-assisted operations, enabling fine-grained, step-wise control through a locking mechanism that preserves prior decisions so subsequent edits affect only specified regions or attributes. Users remain in the loop, making and verifying decisions across stages, while the system applies diffs instead of regenerating full images, reducing drift as fidelity increases. A comparative study with a one-shot baseline shows that participants felt stronger ownership over Creo outputs, as they were able to trace their decisions in building up the image. Furthermore, embedding-based analysis indicates that Creo outputs are less homogeneous than one-shot results. These findings suggest that multi-stage generation, combined with intermediate control and decision locking, is a key design principle for improving controllability, user agency, creativity, and output diversity in generative systems.

Zoe De Simone, Angie Boggust, Arvind Satyanarayan et al. · mit

Generative text-to-image (TTI) models produce high-quality images from short textual descriptions and are widely used in academic and creative domains. Like humans, TTI models have a worldview, a conception of the world learned from their training data and task that influences the images they generate for a given prompt. However, the worldviews of TTI models are often hidden from users, making it challenging for users to build intuition about TTI outputs, and they are often misaligned with users' worldviews, resulting in output images that do not match user expectations. In response, we introduce DiffusionWorldViewer, an interactive interface that exposes a TTI model's worldview across output demographics and provides editing tools for aligning output images with user perspectives. In a user study with 18 diverse TTI users, we find that DiffusionWorldViewer helps users represent their varied viewpoints in generated images and challenge the limited worldview reflected in current TTI models.

Graham M. Jones, Shai Satran, Arvind Satyanarayan

This article proposes a new integration of linguistic anthropology and machine learning (ML) around convergent interests in both the underpinnings of language and making language technologies more socially responsible. While linguistic anthropology focuses on interpreting the cultural basis for human language use, the ML field of interpretability is concerned with uncovering the patterns that Large Language Models (LLMs) learn from human verbal behavior. Through the analysis of a conversation between a human user and an LLM-powered chatbot, we demonstrate the theoretical feasibility of a new, conjoint field of inquiry, cultural interpretability (CI). By focusing attention on the communicative competence involved in the way human users and AI chatbots co-produce meaning in the articulatory interface of human-computer interaction, CI emphasizes how the dynamic relationship between language and culture makes contextually sensitive, open-ended conversation possible. We suggest that, by examining how LLMs internally "represent" relationships between language and culture, CI can: (1) provide insight into long-standing linguistic anthropological questions about the patterning of those relationships; and (2) aid model developers and interface designers in improving value alignment between language models and stylistically diverse speakers and culturally diverse speech communities. Our discussion proposes three critical research axes: relativity, variation, and indexicality.

Angie Boggust, Donghao Ren, Yannick Assogba et al. · apple-ml, cmu

Automated interpretability aims to translate large language model (LLM) features into human understandable descriptions. However, these natural language feature descriptions are often vague, inconsistent, and require manual relabeling. In response, we introduce semantic regexes, structured language descriptions of LLM features. By combining primitives that capture linguistic and semantic feature patterns with modifiers for contextualization, composition, and quantification, semantic regexes produce precise and expressive feature descriptions. Across quantitative benchmarks and qualitative analyses, we find that semantic regexes match the accuracy of natural language while yielding more concise and consistent feature descriptions. Moreover, their inherent structure affords new types of analyses, including quantifying feature complexity across layers, scaling automated interpretability from insights into individual features to model-wide patterns. Finally, in user studies, we find that semantic regex descriptions help people build accurate mental models of LLM feature activations.

Hussein Mozannar, Arvind Satyanarayan, David Sontag

Expert decision makers are starting to rely on data-driven automated agents to assist them with various tasks. For this collaboration to perform properly, the human decision maker must have a mental model of when and when not to rely on the agent. In this work, we aim to ensure that human decision makers learn a valid mental model of the agent's strengths and weaknesses. To accomplish this goal, we propose an exemplar-based teaching strategy where humans solve the task with the help of the agent and try to formulate a set of guidelines of when and when not to defer. We present a novel parameterization of the human's mental model of the AI that applies a nearest neighbor rule in local regions surrounding the teaching examples. Using this model, we derive a near-optimal strategy for selecting a representative teaching set. We validate the benefits of our teaching strategy on a multi-hop question answering task using crowd workers and find that when workers draw the right lessons from the teaching stage, their task performance improves, we furthermore validate our method on a set of synthetic experiments.

Hendrik Strobelt, Benjamin Hoover, Arvind Satyanarayan et al.

While different language models are ubiquitous in NLP, it is hard to contrast their outputs and identify which contexts one can handle better than the other. To address this question, we introduce LMdiff, a tool that visually compares probability distributions of two models that differ, e.g., through finetuning, distillation, or simply training with different parameter sizes. LMdiff allows the generation of hypotheses about model behavior by investigating text instances token by token and further assists in choosing these interesting text instances by identifying the most interesting phrases from large corpora. We showcase the applicability of LMdiff for hypothesis generation across multiple case studies. A demo is available at http://lmdiff.net .

Alan Lundgard, Arvind Satyanarayan

Natural language descriptions sometimes accompany visualizations to better communicate and contextualize their insights, and to improve their accessibility for readers with disabilities. However, it is difficult to evaluate the usefulness of these descriptions, and how effectively they improve access to meaningful information, because we have little understanding of the semantic content they convey, and how different readers receive this content. In response, we introduce a conceptual model for the semantic content conveyed by natural language descriptions of visualizations. Developed through a grounded theory analysis of 2,147 sentences, our model spans four levels of semantic content: enumerating visualization construction properties (e.g., marks and encodings); reporting statistical concepts and relations (e.g., extrema and correlations); identifying perceptual and cognitive phenomena (e.g., complex trends and patterns); and elucidating domain-specific insights (e.g., social and political context). To demonstrate how our model can be applied to evaluate the effectiveness of visualization descriptions, we conduct a mixed-methods evaluation with 30 blind and 90 sighted readers, and find that these reader groups differ significantly on which semantic content they rank as most useful. Together, our model and findings suggest that access to meaningful information is strongly reader-specific, and that research in automatic visualization captioning should orient toward descriptions that more richly communicate overall trends and statistics, sensitive to reader preferences. Our work further opens a space of research on natural language as a data interface coequal with visualization.

Angie Boggust, Benjamin Hoover, Arvind Satyanarayan et al.

Saliency methods -- techniques to identify the importance of input features on a model's output -- are a common step in understanding neural network behavior. However, interpreting saliency requires tedious manual inspection to identify and aggregate patterns in model behavior, resulting in ad hoc or cherry-picked analysis. To address these concerns, we present Shared Interest: metrics for comparing model reasoning (via saliency) to human reasoning (via ground truth annotations). By providing quantitative descriptors, Shared Interest enables ranking, sorting, and aggregating inputs, thereby facilitating large-scale systematic analysis of model behavior. We use Shared Interest to identify eight recurring patterns in model behavior, such as cases where contextual features or a subset of ground truth features are most important to the model. Working with representative real-world users, we show how Shared Interest can be used to decide if a model is trustworthy, uncover issues missed in manual analyses, and enable interactive probing.

Ariel Levy, Monica Agrawal, Arvind Satyanarayan et al.

Automated decision support can accelerate tedious tasks as users can focus their attention where it is needed most. However, a key concern is whether users overly trust or cede agency to automation. In this paper, we investigate the effects of introducing automation to annotating clinical texts--a multi-step, error-prone task of identifying clinical concepts (e.g., procedures) in medical notes, and mapping them to labels in a large ontology. We consider two forms of decision aid: recommending which labels to map concepts to, and pre-populating annotation suggestions. Through laboratory studies, we find that 18 clinicians generally build intuition of when to rely on automation and when to exercise their own judgement. However, when presented with fully pre-populated suggestions, these expert users exhibit less agency: accepting improper mentions, and taking less initiative in creating additional annotations. Our findings inform how systems and algorithms should be designed to mitigate the observed issues.

Harini Suresh, Kathleen M. Lewis, John V. Guttag et al.

Interpretability methods aim to help users build trust in and understand the capabilities of machine learning models. However, existing approaches often rely on abstract, complex visualizations that poorly map to the task at hand or require non-trivial ML expertise to interpret. Here, we present two visual analytics modules that facilitate an intuitive assessment of model reliability. To help users better characterize and reason about a model's uncertainty, we visualize raw and aggregate information about a given input's nearest neighbors. Using an interactive editor, users can manipulate this input in semantically-meaningful ways, determine the effect on the output, and compare against their prior expectations. We evaluate our interface using an electrocardiogram beat classification case study. Compared to a baseline feature importance interface, we find that 14 physicians are better able to align the model's uncertainty with domain-relevant factors and build intuition about its capabilities and limitations.

Harini Suresh, Steven R. Gomez, Kevin K. Nam et al.

To ensure accountability and mitigate harm, it is critical that diverse stakeholders can interrogate black-box automated systems and find information that is understandable, relevant, and useful to them. In this paper, we eschew prior expertise- and role-based categorizations of interpretability stakeholders in favor of a more granular framework that decouples stakeholders' knowledge from their interpretability needs. We characterize stakeholders by their formal, instrumental, and personal knowledge and how it manifests in the contexts of machine learning, the data domain, and the general milieu. We additionally distill a hierarchical typology of stakeholder needs that distinguishes higher-level domain goals from lower-level interpretability tasks. In assessing the descriptive, evaluative, and generative powers of our framework, we find our more nuanced treatment of stakeholders reveals gaps and opportunities in the interpretability literature, adds precision to the design and comparison of user studies, and facilitates a more reflexive approach to conducting this research.

Crystal Lee, Tanya Yang, Gabrielle Inchoco et al.

Controversial understandings of the coronavirus pandemic have turned data visualizations into a battleground. Defying public health officials, coronavirus skeptics on US social media spent much of 2020 creating data visualizations showing that the government's pandemic response was excessive and that the crisis was over. This paper investigates how pandemic visualizations circulated on social media, and shows that people who mistrust the scientific establishment often deploy the same rhetorics of data-driven decision-making used by experts, but to advocate for radical policy changes. Using a quantitative analysis of how visualizations spread on Twitter and an ethnographic approach to analyzing conversations about COVID data on Facebook, we document an epistemological gap that leads pro- and anti-mask groups to draw drastically different inferences from similar data. Ultimately, we argue that the deployment of COVID data visualizations reflect a deeper sociopolitical rift regarding the place of science in public life.

Rupayan Neogy, Jonathan Zong, Arvind Satyanarayan

Establishing common ground and maintaining shared awareness amongst participants is a key challenge in collaborative visualization. For real-time collaboration, existing work has primarily focused on synchronizing constituent visualizations - an approach that makes it difficult for users to work independently, or selectively attend to their collaborators' activity. To address this gap, we introduce a design space for representing synchronous multi-user collaboration in visualizations defined by two orthogonal axes: situatedness, or whether collaborators' interactions are overlaid on or shown outside of a user's view, and specificity, or whether collaborators are depicted through abstract, generic representations or through specific means customized for the given visualization. We populate this design space with a variety of examples including generic and custom synchronized cursors, and user legends that collect these cursors together or reproduce collaborators' views as thumbnails. To build common ground, users can interact with these representations by peeking to take a quick look at a collaborator's view, tracking to follow along with a collaborator in real-time, and forking to independently explore the visualization based on a collaborator's work. We present a reference implementation of a wrapper library that converts interactive Vega-Lite charts into collaborative visualizations. We find that our approach affords synchronous collaboration across an expressive range of visual designs and interaction techniques.

Jonathan Zong, Dhiraj Barnwal, Rupayan Neogy et al.

Recent graphical interfaces offer direct manipulation mechanisms for authoring visualizations, but are largely restricted to static output. To author interactive visualizations, users must instead turn to textual specification; but, such approaches impose a higher technical burden than their graphical counterparts. To bridge this gap, we introduce interaction design by demonstration: a novel method for authoring interaction techniques via direct manipulation. Users perform an interaction (e.g., button clicks, drags, or key presses) directly on the visualization they are editing. The system interprets this performance using a set of heuristics, and produces suggestions of possible interaction designs. Heuristics account for properties of the interaction (e.g., target and event type) as well as the visualization (e.g., mark and scale types, and multiple views). Interaction design suggestions are displayed as thumbnails; users can preview and test these suggestions, iteratively refine them through additional demonstrations, and finally apply and customize them via property inspectors. To evaluate our approach, we instantiate it in Lyra, an existing visualization design environment. We demonstrate its expressive extent with a gallery of diverse examples, and evaluate its usability through a first-use study and via an analysis of its cognitive dimensions. We find that, in Lyra, interaction design by demonstration enables users to rapidly express a wide range of interactive visualizations.

Nava Haghighi, Nathalie Vladis, Yuanbo Liu et al.

With the rise of wearables, haptic interfaces are increasingly favored to communicate information in an ambient manner. Despite this expectation, existing guidelines are developed in studies where the participant's focus is entirely on the haptic task. In this work, we systematically study the cognitive load imposed by properties of a haptic signal. Participants wear a haptic device on their forearm, and are asked to perform a 1-back task. Each experimental condition isolates an individual property of the haptic signal (e.g., amplitude, waveform, rhythm) and participants are asked to identify the gradient of the data. We evaluate each condition across 16 participants, measuring participants' response times, error rates, and qualitative and quantitative surveys (e.g., NASA TLX). Our results indicate that gender and language differences may impact preference for some properties, that participants prefer properties that can be rapidly identified, and that amplitude imposes the lowest cognitive load.

Angie Boggust, Brandon Carter, Arvind Satyanarayan

Embeddings mapping high-dimensional discrete input to lower-dimensional continuous vector spaces have been widely adopted in machine learning applications as a way to capture domain semantics. Interviewing 13 embedding users across disciplines, we find comparing embeddings is a key task for deployment or downstream analysis but unfolds in a tedious fashion that poorly supports systematic exploration. In response, we present the Embedding Comparator, an interactive system that presents a global comparison of embedding spaces alongside fine-grained inspection of local neighborhoods. It systematically surfaces points of comparison by computing the similarity of the $k$-nearest neighbors of every embedded object between a pair of spaces. Through case studies across multiple modalities, we demonstrate our system rapidly reveals insights, such as semantic changes following fine-tuning, language changes over time, and differences between seemingly similar models. In evaluations with 15 participants, we find our system accelerates comparisons by shifting from laborious manual specification to browsing and manipulating visualizations.

Arvind Satyanarayan, Bongshin Lee, Donghao Ren et al.

An emerging generation of visualization authoring systems support expressive information visualization without textual programming. As they vary in their visualization models, system architectures, and user interfaces, it is challenging to directly compare these systems using traditional evaluative methods. Recognizing the value of contextualizing our decisions in the broader design space, we present critical reflections on three systems we developed -- Lyra, Data Illustrator, and Charticulator. This paper surfaces knowledge that would have been daunting within the constituent papers of these three systems. We compare and contrast their (previously unmentioned) limitations and trade-offs between expressivity and learnability. We also reflect on common assumptions that we made during the development of our systems, thereby informing future research directions in visualization authoring systems.

Yifan Wu, Remco Chang, Joseph Hellerstein et al.

Interactive visualization design and research have primarily focused on local data and synchronous events. However, for more complex use cases---e.g., remote database access and streaming data sources---developers must grapple with distributed data and asynchronous events. Currently, constructing these use cases is difficult and time-consuming; developers are forced to operationally program low-level details like asynchronous database querying and reactive event handling. This approach is in stark contrast to modern methods for browser-based interactive visualization, which feature high-level declarative specifications. In response, we present DIEL, a declarative framework that supports asynchronous events over distributed data. Like many declarative visualization languages, DIEL developers need only specify what data they want, rather than procedural steps for how to assemble it; uniquely, DIEL models asynchronous events (e.g., user interactions or server responses) as streams of data that are captured in event logs. To specify the state of a user interface at any time, developers author declarative queries over the data and event logs; DIEL compiles and optimizes a corresponding dataflow graph, and synthesizes necessary low-level distributed systems details. We demonstrate DIEL's performance and expressivity through ex-ample interactive visualizations that make diverse use of remote data and coordination of asynchronous events. We further evaluate DIEL's usability using the Cognitive Dimensions of Notations framework, revealing wins such as ease of change, and compromises such as premature commitments.

Madelon Hulsebos, Kevin Hu, Michiel Bakker et al.

Correctly detecting the semantic type of data columns is crucial for data science tasks such as automated data cleaning, schema matching, and data discovery. Existing data preparation and analysis systems rely on dictionary lookups and regular expression matching to detect semantic types. However, these matching-based approaches often are not robust to dirty data and only detect a limited number of types. We introduce Sherlock, a multi-input deep neural network for detecting semantic types. We train Sherlock on $686,765$ data columns retrieved from the VizNet corpus by matching $78$ semantic types from DBpedia to column headers. We characterize each matched column with $1,588$ features describing the statistical properties, character distributions, word embeddings, and paragraph vectors of column values. Sherlock achieves a support-weighted F$_1$ score of $0.89$, exceeding that of machine learning baselines, dictionary and regular expression benchmarks, and the consensus of crowdsourced annotations.

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.