Analyzing literature involves tracking interactions between characters, locations, and themes. Visualization has the potential to facilitate the mapping and analysis of these complex relationships, but capturing structured information from unstructured story data remains a challenge. As large language models (LLMs) continue to advance, we see an opportunity to use their text processing and analysis capabilities to augment and reimagine existing storyline visualization techniques. Toward this goal, we introduce an LLM-driven data parsing pipeline that automatically extracts relevant narrative information from novels and scripts. We then apply this pipeline to create Story Ribbons, an interactive visualization system that helps novice and expert literary analysts explore detailed character and theme trajectories at multiple narrative levels. Through pipeline evaluations and user studies with Story Ribbons on 36 literary works, we demonstrate the potential of LLMs to streamline narrative visualization creation and reveal new insights about familiar stories. We also describe current limitations of AI-based systems, and interaction motifs designed to address these issues.

In large language model (LLM) pretraining, data quality is believed to determine model quality. In this paper, we re-examine the notion of "quality" from the perspective of pre- and post-training co-design. Specifically, we explore the possibility that pre-training on more toxic data can lead to better control in post-training, ultimately decreasing a model's output toxicity. First, we use a toy experiment to study how data composition affects the geometry of features in the representation space. Next, through controlled experiments with Olmo-1B models trained on varying ratios of clean and toxic data, we find that the concept of toxicity enjoys a less entangled linear representation as the proportion of toxic data increases. Furthermore, we show that although toxic data increases the generational toxicity of the base model, it also makes the toxicity easier to remove. Evaluations on Toxigen and Real Toxicity Prompts demonstrate that models trained on toxic data achieve a better trade-off between reducing generational toxicity and preserving general capabilities when detoxifying techniques such as inference-time intervention (ITI) are applied. Our findings suggest that, with post-training taken into account, bad data may lead to good models.

Visualizing high dimensional data is challenging, since any dimensionality reduction technique will distort distances. A classic method in cartography–Tissot’s Indicatrix, specific to sphere-to-plane maps– visualizes distortion using ellipses. Inspired by this idea, we describe the hypertrix: a method for representing distortions that occur when data is projected from arbitrarily high dimensions onto a 2D plane. We demonstrate our technique through synthetic and real-world datasets, and describe how this indicatrix can guide interpretations of nonlinear dimensionality reduction.

Transformer models are revolutionizing machine learning, but their inner workings remain mysterious. In this work, we present a new visualization technique designed to help researchers understand the self-attention mechanism in transformers that allows these models to learn rich, contextual relationships between elements of a sequence. The main idea behind our method is to visualize a joint embedding of the query and key vectors used by transformer models to compute attention. Unlike previous attention visualization techniques, our approach enables the analysis of global patterns across multiple input sequences. We create an interactive visualization tool, AttentionViz (demo: http://attentionviz.com), based on these joint query-key embeddings, and use it to study attention mechanisms in both language and vision transformers. We demonstrate the utility of our approach in improving model understanding and offering new insights about query-key interactions through several application scenarios and expert feedback.

Artificial intelligence (AI) presents new challenges for the user experience (UX) of products and services. Recently, practitioner-facing resources and design guidelines have become available to ease some of these challenges. However, little research has investigated if and how these guidelines are used, and how they impact practice. In this paper, we investigated how industry practitioners use the People + AI Guidebook. We conducted interviews with 31 practitioners (i.e., designers, product managers) to understand how they use human-AI guidelines when designing AI-enabled products. Our findings revealed that practitioners use the guidebook not only for addressing AI’s design challenges, but also for education, cross-functional communication, and for developing internal resources. We uncovered that practitioners desire more support for early phase ideation and problem formulation to avoid AI product failures. We discuss the implications for future resources aiming to help practitioners in designing AI products.

We introduce Inference-Time Intervention (ITI), a technique designed to enhance the "truthfulness" of large language models (LLMs). ITI operates by shifting model activations during inference, following a learned set of directions across a limited number of attention heads. This intervention significantly improves the performance of LLaMA models on the TruthfulQA benchmark. On an instruction-finetuned LLaMA called Alpaca, ITI improves its truthfulness from $32.5\%$ to $65.1\%$. We identify a tradeoff between truthfulness and helpfulness and demonstrate how to balance it by tuning the intervention strength. ITI is minimally invasive and computationally inexpensive. Moreover, the technique is data efficient: while approaches like RLHF require extensive annotations, ITI locates truthful directions using only few hundred examples. Our findings suggest that LLMs may have an internal representation of the likelihood of something being true, even as they produce falsehoods on the surface.

A key challenge in developing and deploying Machine Learning (ML) systems is understanding their performance across a wide range of inputs. To address this challenge, we created the What-If Tool, an open-source application that allows practitioners to probe, visualize, and analyze ML systems, with minimal coding. The What-If Tool lets practitioners test performance in hypothetical situations, analyze the importance of different data features, and visualize model behavior across multiple models and subsets of input data. It also lets practitioners measure systems according to multiple ML fairness metrics. We describe the design of the tool, and report on real-life usage at different organizations.

Machine learning (ML) is increasingly being used in image retrieval systems for medical decision making. One application of ML is to retrieve visually similar medical images from past patients (e.g. tissue from biopsies) to reference when making a medical decision with a new patient. However, no algorithm can perfectly capture an expert's ideal notion of similarity for every case: an image that is algorithmically determined to be similar may not be medically relevant to a doctor's specific diagnostic needs. In this paper, we identified the needs of pathologists when searching for similar images retrieved using a deep learning algorithm, and developed tools that empower users to cope with the search algorithm on-the-fly, communicating what types of similarity are most important at different moments in time. In two evaluations with pathologists, we found that these tools increased the diagnostic utility of images found and increased user trust in the algorithm. The tools were preferred over a traditional interface, without a loss in diagnostic accuracy. We also observed that users adopted new strategies when using refinement tools, re-purposing them to test and understand the underlying algorithm and to disambiguate ML errors from their own errors. Taken together, these findings inform future human-ML collaborative systems for expert decision-making.

Saliency methods can aid understanding of deep neural networks. Recent years have witnessed many improvements to saliency methods, as well as new ways for evaluating them. In this paper, we 1) present a novel region-based attribution method, XRAI, that builds upon integrated gradients (Sundararajan et al. 2017), 2) introduce evaluation methods for empirically assessing the quality of image-based saliency maps (Performance Information Curves (PICs)), and 3) contribute an axiom-based sanity check for attribution methods. Through empirical experiments and example results, we show that XRAI produces better results than other saliency methods for common models and the ImageNet dataset.

Transformer architectures show significant promise for natural language processing. Given that a single pretrained model can be fine-tuned to perform well on many different tasks, these networks appear to extract generally useful linguistic features. A natural question is how such networks represent this information internally. This paper describes qualitative and quantitative investigations of one particularly effective model, BERT. At a high level, linguistic features seem to be represented in separate semantic and syntactic subspaces. We find evidence of a fine-grained geometric representation of word senses. We also present empirical descriptions of syntactic representations in both attention matrices and individual word embeddings, as well as a mathematical argument to explain the geometry of these representations.

Recent success in deep learning has generated immense interest among practitioners and students, inspiring many to learn about this new technology. While visual and interactive approaches have been successfully developed to help people more easily learn deep learning, most existing tools focus on simpler models. In this work, we present GAN Lab, the first interactive visualization tool designed for non-experts to learn and experiment with Generative Adversarial Networks (GANs), a popular class of complex deep learning models. With GAN Lab, users can interactively train generative models and visualize the dynamic training process's intermediate results. GAN Lab tightly integrates an model overview graph that summarizes GAN's structure, and a layered distributions view that helps users interpret the interplay between submodels. GAN Lab introduces new interactive experimentation features for learning complex deep learning models, such as step-by-step training at multiple levels of abstraction for understanding intricate training dynamics. Implemented using TensorFlow.js, GAN Lab is accessible to anyone via modern web browsers, without the need for installation or specialized hardware, overcoming a major practical challenge in deploying interactive tools for deep learning.

The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net’s internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result–for example, how sensitive a prediction of “zebra” is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.

We present a design study of the TensorFlow Graph Visualizer, part of the TensorFlow machine intelligence platform. This tool helps users understand complex machine learning architectures by visualizing their underlying dataflow graphs. The tool works by applying a series of graph transformations that enable standard layout techniques to produce a legible interactive diagram. To declutter the graph, we decouple non-critical nodes from the layout. To provide an overview, we build a clustered graph using the hierarchical structure annotated in the source code. To support exploration of nested structure on demand, we perform edge bundling to enable stable and responsive cluster expansion. Finally, we detect and highlight repeated structures to emphasize a model's modular composition. To demonstrate the utility of the visualizer, we describe example usage scenarios and report user feedback. Overall, users find the visualizer useful for understanding, debugging, and sharing the structures of their models.

We propose a simple solution to use a single Neural Machine Translation (NMT) model to translate between multiple languages. Our solution requires no changes to the model architecture from a standard NMT system but instead introduces an artificial token at the beginning of the input sentence to specify the required target language. Using a shared wordpiece vocabulary, our approach enables Multilingual NMT systems using a single model. On the WMT’14 benchmarks, a single multilingual model achieves comparable performance for English→French and surpasses state-of-theart results for English→German. Similarly, a single multilingual model surpasses state-of-the-art results for French→English and German→English on WMT’14 and WMT’15 benchmarks, respectively. On production corpora, multilingual models of up to twelve language pairs allow for better translation of many individual pairs. Our models can also learn to perform implicit bridging between language pairs never seen explicitly during training, showing that transfer learning and zero-shot translation is possible for neural translation. Finally, we show analyses that hints at a universal interlingua representation in our models and also show some interesting examples when mixing languages.

G+ Ripples is a visualization of information flow that shows users how public posts are shared on Google+. Unlike other social network visualizations, Ripples exists as a "native" visualization: it is directly accessible from public posts on Google+. This unique position leads to both new constraints and new possibilities for design. We describe the visualization technique, which is a new mix of node-and-link and circular treemap metaphors. We then describe user reactions as well as some of the patterns of sharing that are made evident by the Ripples visualization.

Data visualization has historically been accessible only to the elite in academia, business, and government. It was "serious" technology, created by experts for experts. In recent years, however, web-based visualizations--ranging from political art projects to news stories--have reached audiences of millions. What will this new era of data transparency look like--and what are the implications for technologists who work with data? To help answer this question, we report on recent research into public data analysis and visualization. Some of our results come from Many Eyes, a "living laboratory" web site where people may upload their own data, create interactive visualizations, and carry on conversations. We'll also show how the art world has embraced visualization. We'll discuss the future of visual literacy and what it means for a world where visualizations are a part of political discussions, citizen activism, religious discussions, game playing, and educational exchanges.

Data visualization has historically been accessible only to the elite in academia, business, and government. It was "serious" technology, created by experts for experts. In recent years, however, web-based visualizations--ranging from political art projects to news stories--have reached audiences of millions. Meanwhile, new initiatives in government, aimed at all citizens, point to an era of increased transparency.

We present a new technique, the phrase net, for generating visual overviews of unstructured text. A phrase net displays a graph whose nodes are words and whose edges indicate that two words are linked by a user-specified relation. These relations may be defined either at the syntactic or lexical level; different relations often produce very different perspectives on the same text. Taken together, these perspectives often provide an illuminating visual overview of the key concepts and relations in a document or set of documents.

We discuss the design and usage of "Wordle," a web-based tool for visualizing text. Wordle creates tag-cloud-like displays that give careful attention to typography, color, and composition. We describe the algorithms used to balance various aesthetic criteria and create the distinctive Wordle layouts. We then present the results of a study of Wordle usage, based both on spontaneous behaviour observed in the wild, and on a large-scale survey of Wordle users. The results suggest that Wordles have become a kind of medium of expression, and that a "participatory culture" has arisen around them.

Many Eyes is a web site that provides collaborative visualization services, allowing users to upload data sets, visualize them, and comment on each other's visualizations. This paper describes a first interview-based study of Many Eyes users, which sheds light on user motivation for creating public visualizations. Users talked about data for many reasons, from scientific research to political advocacy to hobbies. One consistent theme across these different scenarios is the use of visualizations in communication and collaborative practices. Collaboration and conversation, however, often took place outside the site, leaving no traces on Many Eyes itself. In other words, despite spurring significant social activity, Many Eyes is not so much an online community as a "community component" which users insert into pre-existing online social systems.

We introduce the Word Tree, a new visualization and information-retrieval technique aimed at text documents. A word tree is a graphical version of the traditional "keyword-in-context" method, and enables rapid querying and exploration of bodies of text. In this paper we describe the design of the technique, along with some of the technical issues that arise in its implementation. In addition, we discuss the results of several months of public deployment of word trees on Many Eyes, which provides a window onto the ways in which users obtain value from the visualization.

This paper describes mechanisms for asynchronous collaboration in the context of information visualization, recasting visualizations as not just analytic tools, but social spaces. We contribute the design and implementation of sense.us, a web site supporting asynchronous collaboration across a variety of visualization types. The site supports view sharing, discussion, graphical annotation, and social navigation and includes novel interaction elements. We report the results of user studies of the system, observing emergent patterns of social data analysis, including cycles of observation and hypothesis, and the complementary roles of social navigation and data-driven exploration.

We describe the design and deployment of Many Eyes, a public web site where users may upload data, create interactive visualizations, and carry on discussions. The goal of the site is to support collaboration around visualizations at a large scale by fostering a social style of data analysis in which visualizations not only serve as a discovery tool for individuals but also as a medium to spur discussion among users. To support this goal, the site includes novel mechanisms for end-user creation of visualizations and asynchronous collaboration around those visualizations. In addition to describing these technologies, we provide a preliminary report on the activity of our users.

We present Themail, a visualization that portrays relationships using the interaction histories preserved in email archives. Using the content of exchanged messages, it shows the words that characterize one's correspondence with an individual and how they change over the period of the relationship.This paper describes the interface and content-parsing algorithms in Themail. It also presents the results from a user study where two main interaction modes with the visualization emerged: exploration of "big picture" trends and themes in email (haystack mode) and more detail-oriented exploration (needle mode). Finally, the paper discusses the limitations of the content parsing approach in Themail and the implications for further research on email content visualization.

The Internet has fostered an unconventional and powerful style of collaboration: "wiki" web sites, where every visitor has the power to become an editor. In this paper we investigate the dynamics of Wikipedia, a prominent, thriving wiki. We make three contributions. First, we introduce a new exploratory data analysis tool, the history flow visualization, which is effective in revealing patterns within the wiki context and which we believe will be useful in other collaborative situations as well. Second, we discuss several collaboration patterns highlighted by this visualization tool and corroborate them with statistical analysis. Third, we discuss the implications of these patterns for the design and governance of online collaborative social spaces. We focus on the relevance of authorship, the value of community surveillance in ameliorating antisocial behavior, and how authors with competing perspectives negotiate their differences.

Although current online chat environments provide new opportunities for communication, they are quite constrained in their ability to convey many important pieces of social information, ranging from the number of participants in a conversation to the subtle nuances of expression that enrich face to face speech. In this paper we present Chat Circles, an abstract graphical interface for synchronous conversa-tion. Here, presence and activity are made manifest by changes in color and form, proximity-based filtering intuitively breaks large groups into conversational clusters, and the archives of a conversation are made visible through an integrated history interface. Our goal in this work is to create a richer environment for online discussions.