Acoustic monitoring is an extremely useful tool for conservationists to track and identify elephants deep within densely wooded environments. However, traditional acoustic audio event detection approaches achieve limited success for two reasons. First, elephant calls are rare in continuous audio data and vary significantly in their duration, frequency, and harmonics. Second, other rare forest noises can easily be mistaken for elephant calls. To tackle class imbalance, we use majority class undersampling and introduce a data augmentation technique that generates additional realistic positive elephant call training examples. Moreover, to handle rare, challenging background sounds, we introduce a novel curriculum driven training strategy that identifies hard to classify background, and uses such sounds to incrementally improve the model. We show that the curriculum approach significantly improves upon our baseline model. Additionally, we introduce the limit case of the curriculum strategy, captured by a new two-stage modeling framework. Our results show that combining these approaches of data augmentation and curriculum-driven learning leads to significant model improvement, achieving a test F1 score of 0.74 compared to 0.45 for the baseline model.

Artificial intelligence is increasingly used in ecological contexts to monitor animal and insect populations. Species of interest are those in danger of extinction, and those that play pivotal roles in agriculture. Noticing population declines or geographical shifts early enough for intervention can prevent local famine and disruption to the global food chain. Traditionally, data are collected in the field using human labor or sensors. Applicable classification models then analyze the data on central servers. The most expensive, and sometimes dangerous part of the remote sensing solution is the human labor of visiting the sensors, retrieving data, and changing batteries. Constantly sending all readings by radio is expensive in power. Instead, having AI in the sensors process readings, and only transmitting results could lead to an indefinitely autonomous, renewably powered solution. We implemented an elephant vocalization detector on a small processor board, and demonstrate that such a device can be operated at low enough power levels with considerable freedom of choice among AI technologies. We achieved a mean of 1.6W, in the best case staying within 75% of memory limits. Measurements covered three inference models, two batch sizes, and two floating point word width settings.

Committing to a major is a fateful step in an undergraduate's education, yet the relationship between courses taken early in an academic career and ultimate major choice remains little studied at scale. We analyze transcript data capturing the academic careers of 26,892 undergraduates at a private university between 2000 and 2020. We forecast students' terminal major on the basis of course-choice sequences beginning at university entry. We represent course enrollment history using natural-language methods and vector embeddings. We find that a student's very first enrolled course predicts their terminal major thirty times better than random guessing and more than a third better than majority class voting.

We explore the relationship between video playback speed and student learning outcomes. Using an experimental design, we present the results of a pre-registered study that assigns users to watch videos at either 1.0x or 1.25x speed. We find that students who consume sped content are more likely to get better grades in a course, attempt more content, and obtain more certificates. We also find that when videos are sped up, students spend less time consuming videos and are marginally more likely to complete more video content. These findings suggest that future study of playback speed as a tool for optimizing video content for MOOCs is warranted. Applications for reinforcement learning and adaptive content are discussed.

The processes through which course selections accumulate into college pathways in US higher education is poorly instrumented for observation at scale. We offer an analytic toolkit, called Via, which transforms commonly available enrollment data into formal graphs that are amenable to interactive visualizations and computational exploration. We explain the procedures required to project enrollment records onto graphs, and then demonstrate the toolkit utilizing eighteen years of enrollment data at a large private research university. Findings complement prior research on academic search and offer powerful new means for making pathway navigation more efficient.

A study deployed the mental health Relational Frame Theory as grounding for an analysis of sentiment dynamics in human-language dialogs. The work takes a step towards enabling use of conversational agents in mental health settings. Sentiment tendencies and mirroring behaviors in 11k human-human dialogs were compared with behaviors when humans interacted with conversational agents in a similar-sized collection. The study finds that human sentiment-related interaction norms persist in human-agent dialogs, but that humans are twice as likely to respond negatively when faced with a negative utterance by a robot than in a comparable situation with humans. Similarly, inhibition towards use of obscenity is greatly reduced. We introduce a new Affective Neural Net implementation that specializes in analyzing sentiment in real time.

The software platforms that mediate online learning experiences are the common ground where learning science and computer science intersect. This panel will discuss the affordances of current online learning platforms and lessons learned in using them with students. The goal of the panel is to help learning scientists and computer scientists understand each others' needs and how they might be effectively addressed in these platforms. The panelists, who have experience creating/using these platforms and interacting with learning scientists, will discuss how current platforms for learning at scale might evolve to better serve the community.

We describe our experience exhibiting a human-size robot in a museum, encouraging visitors to interact with the robot and even program it to perform a sequence of timed poses. At the museum, users' programs were run on a real robot for all to see. The installation attracted and engaged visitors from age two to adult. The most intuitive of our interfaces was equally captivating for young and older visitors. We present the pros and cons of our interfaces and the engagement of visitors at the exhibit as lessons for other exhibitors who aim to achieve active prolonged engagement with robots in museum settings.

Data quality issues such as missing, erroneous, extreme and duplicate values undermine analysis and are time-consuming to find and fix. Automated methods can help identify anomalies, but determining what constitutes an error is context-dependent and so requires human judgment. While visualization tools can facilitate this process, analysts must often manually construct the necessary views, requiring significant expertise. We present Profiler, a visual analysis tool for assessing quality issues in tabular data. Profiler applies data mining methods to automatically flag problematic data and suggests coordinated summary visualizations for assessing the data in context. The system contributes novel methods for integrated statistical and visual analysis, automatic view suggestion, and scalable visual summaries that support real-time interaction with millions of data points. We present Profiler's architecture --- including modular components for custom data types, anomaly detection routines and summary visualizations --- and describe its application to motion picture, natural disaster and water quality data sets.

The Robots for Humanity project aims to enable people with severe motor impairments to interact with their own bodies and their environment through the use of an assistive mobile manipulator, thereby improving their quality of life. Assistive mobile manipulators (AMMs) are mobile robots that physically manipulate the world in order to provide assistance to people with disabilities. They present an exciting frontier for assistive technology, as they can operate away from the user, have a large dexterous workspace (due to their mobility), and not directly encumber their users. The cornerstone of this project is an ongoing, interactive design process with a quadriplegic user, Henry Evans, and his wife and primary caregiver, Jane Evans. Henry has been enabled, through the use of a PR2 robot, to scratch his own face, shave, fetch a towel from his kitchen, and hand out Halloween candy to trick-ortreating children at a local mall.

Organizations rely on data analysts to model customer engagement, streamline operations, improve production, inform business decisions, and combat fraud. Though numerous analysis and visualization tools have been built to improve the scale and efficiency at which analysts can work, there has been little research on how analysis takes place within the social and organizational context of companies. To better understand the enterprise analysts' ecosystem, we conducted semi-structured interviews with 35 data analysts from 25 organizations across a variety of sectors, including healthcare, retail, marketing and finance. Based on our interview data, we characterize the process of industrial data analysis and document how organizational features of an enterprise impact it. We describe recurring pain points, outstanding challenges, and barriers to adoption for visual analytic tools. Finally, we discuss design implications and opportunities for visual analysis research.

Though data analysis tools continue to improve, analysts still expend an inordinate amount of time and effort manipulating data and assessing data quality issues. Such "data wrangling" regularly involves reformatting data values or layout, correcting erroneous or missing values, and integrating multiple data sources. These transforms are often difficult to specify and difficult to reuse across analysis tasks, teams, and tools. In response, we introduce Wrangler, an interactive system for creating data transformations. Wrangler combines direct manipulation of visualized data with automatic inference of relevant transforms, enabling analysts to iteratively explore the space of applicable operations and preview their effects. Wrangler leverages semantic data types (e.g., geographic locations, dates, classification codes) to aid validation and type conversion. Interactive histories support review, refinement, and annotation of transformation scripts. User study results show that Wrangler significantly reduces specification time and promotes the use of robust, auditable transforms instead of manual editing.

In our field deployments of mobile remote presence (MRP) systems in offices, we observed that remote operators of MRPs often unintentionally spoke too loudly. This disrupted their local co-workers, who happened to be within earshot of the MRP system. To address this issue, we prototyped and empirically evaluated the effect of sidetone to help operators self regulate their speaking loudness. Sidetone is the intentional, attenuated feedback of speakers' voices to their ears while they are using a telecommunication device. In a 3-level (no sidetone vs. low sidetone vs. high sidetone) within- participants pair of experiments, people interacted with a confederate through an MRP system. The first experiment involved MRP operators using headsets with boom microphones (N=20). The second experiment involved MRP operators using loudspeakers and desktop microphones (N=14). While we detected the effects of the sidetone manipulation in our audio-visual context, the effect was attenuated in comparison to earlier audio-only studies. We hypothesize that the strong visual component of our MRP system interferes with the sidetone effect. We also found that engaging in more social tasks (e.g., a getting-to-know-you activity) and more intellectually demanding tasks (e.g., a creativity exercise) influenced how loudly people spoke. This suggests that testing such sidetone effects in the typical read-aloud setting is insufficient for generalizing to more interactive, communication tasks. We conclude that MRP application support must reach beyond the time honored audio-only technologies to solve the problem of excessive speaker loudness.

A fundamental premise of tagging systems is that regular users can organize large collections for browsing and other tasks using uncontrolled vocabularies. Until now, that premise has remained relatively unexamined. Using library data, we test the tagging approach to organizing a collection. We find that tagging systems have three major large scale organizational features: consistency, quality, and completeness. In addition to testing these features, we present results suggesting that users produce tags similar to the topics designed by experts, that paid tagging can effectively supplement tags in a tagging system, and that information integration may be possible across tagging systems.

PhotoSpread is a spreadsheet system for organizing and analyzing photo collections. It extends the current spreadsheet paradigm in two ways: (a) PhotoSpread accommodates sets of objects (e.g., photos) annotated with tags (attribute-value pairs). Formulas can manipulate object sets and refer to tags. (b) Photos can be reorganized (tags and location changed) by drag-and-drop operations on the spreadsheet. The PhotoSpread design was driven by the needs of field biologists who have large collections of annotated photos. The paper describes the PhotoSpread functionality and the design choices made.

We developed and tested a heuristic technique for extracting the main article from news site Web pages. We construct the DOM tree of the page and score every node based on the amount of text, the number of links it contains and additional heuristics. The method is site-independent and does not use any language-based features. We tested our algorithm on a set of 1120 news article pages from 27 domains. Our algorithm achieved over 97% precision and 98% recall, and an average processing speed of under 15ms per page.

Using gaze information as a form of input poses challenges based on the nature of eye movements and how we humans use our eyes in conjunction with other motor actions. In this paper, we present three techniques for improving the use of gaze as a form of input. We first present a saccade detection and smoothing algorithm that works on real-time streaming gaze information. We then present a study which explores some of the timing issues of using gaze in conjunction with a trigger (key press or other motor action) and propose a solution for resolving these issues. Finally, we present the concept of Focus Points, which makes it easier for users to focus their gaze when using gaze-based interaction techniques. Though these techniques were developed for improving the performance of gaze-based pointing, their use is applicable in general to using gaze as a practical form of input.

Motivated by our work with political scientists who need to manually analyze large Web archives of news sites, we present SpotSigs, a new algorithm for extracting and matching signatures for near duplicate detection in large Web crawls. Our spot signatures are designed to favor natural-language portions of Web pages over advertisements and navigational bars.

This paper explores architectural support for interfaces combining pen, paper, and PC. We show how the eventbased approach common to GUIs can apply to augmented paper, and describe additions to address paper’s distinguishing characteristics. To understand the developer experience of this architecture, we deployed the toolkit to 17 student teams for six weeks. Analysis of the developers ’ code provided insight into the appropriateness of events for paper UIs. The usage patterns we distilled informed a second iteration of the toolkit, which introduces techniques for integrating interactive and batched input handling, coordinating interactions across devices, and debugging paper applications. The study also revealed that programmers created gesture handlers by composing simple ink measurements. This desire for informal interactions inspired us to include abstractions for recognition. This work has implications beyond paper — designers of graphical tools can examine API usage to inform iterative toolkit development.

BACKGROUND: We sketch our species identification tool for palm sized computers that helps knowledgeable observers with census activities. An algorithm turns an identification matrix into a minimal length series of questions that guide the operator towards identification. Historic observation data from the census geographic area helps minimize question volume. We explore how much historic data is required to boost performance, and whether the use of history negatively impacts identification of rare species. We also explore how characteristics of the matrix interact with the algorithm, and how best to predict the probability of observing a previously unseen species. RESULTS: Point counts of birds taken at Stanford University's Jasper Ridge Biological Preserve between 2000 and 2005 were used to examine the algorithm. A computer identified species by correctly answering, and counting the algorithm's questions. We also explored how the character density of the key matrix and the theoretical minimum number of questions for each bird in the matrix influenced the algorithm. Our investigation of the required probability smoothing determined whether Laplace smoothing of observation probabilities was sufficient, or whether the more complex Good-Turing technique is required. CONCLUSION: Historic data improved identification speed, but only impacted the top 25% most frequently observed birds. For rare birds the history based algorithms did not impose a noticeable penalty in the number of questions required for identification. For our dataset neither age of the historic data, nor the number of observation years impacted the algorithm. Density of characters for different taxa in the identification matrix did not impact the algorithms. Intrinsic differences in identifying different birds did affect the algorithm, but the differences affected the baseline method of not using historic data to exactly the same degree. We found that Laplace smoothing performed better for rare species than Simple Good-Turing, and that, contrary to expectation, the technique did not then adversely affect identification performance for frequently observed birds.

We present a practical technique for pointing and selection using a combination of eye gaze and keyboard triggers. EyePoint uses a two-step progressive refinement process fluidly stitched together in a look-press-look-release action, which makes it possible to compensate for the accuracy limitations of the current state-of-the-art eye gaze trackers. While research in gaze-based pointing has traditionally focused on disabled users, EyePoint makes gaze-based pointing effective and simple enough for even able-bodied users to use for their everyday computing tasks. As the cost of eye gaze tracking devices decreases, it will become possible for such gaze-based techniques to be used as a viable alternative for users who choose not to use a mouse depending on their abilities, tasks and preferences.

This paper describes a communication-minded visualization called progressive multiples that supports both the forensic analysis and presentation of multidimensional event data. We combine ideas from progressive disclosure, which reveals data to the user on demand, and small multiples [21], which allows users to compare many images at once. Sets of events are visualized as timelines. Events are placed in temporal order on the x-axis, and a scalar dimension of the data is mapped to the y-axis. To support forensic analysis, users can pivot from an event in an existing timeline to create a new timeline of related events. The timelines serve as an exploration history, which has two benefits. First, this exploration history allows users to backtrack and explore multiple paths. Second, once a user has concluded an analysis, these timelines serve as the raw visual material for composing a story about the analysis. A narrative that conveys the analytical result can be created for a third party by copying and reordering timelines from the history. Our work is motivated by working with network security administrators and researchers in political communication. We describe a prototype that we are deploying with administrators and the results of a user study where we applied our technique to the visualization of a simulated epidemic.

Multi-user, touch-sensing input devices create opportunities for the use of cooperative gestures -- multi-user gestural interactions for single display groupware. Cooperative gestures are interactions where the system interprets the gestures of more than one user as contributing to a single, combined command. Cooperative gestures can be used to enhance users' sense of teamwork, increase awareness of important system events, facilitate reachability and access control on large, shared displays, or add a unique touch to an entertainment-oriented activity. This paper discusses motivating scenarios for the use of cooperative gesturing and describes some initial experiences with CollabDraw, a system for collaborative art and photo manipulation. We identify design issues relevant to cooperative gesturing interfaces, and present a preliminary design framework. We conclude by identifying directions for future research on cooperative gesturing interaction techniques.

We explore how the placement of control widgets (such as menus) affects collaboration and usability for co-located tabletop groupware applications. We evaluated two design alternatives: a centralized set of controls shared by all users, and separate per-user controls replicated around the borders of the shared tabletop. We conducted this evaluation in the context of TeamTag, a system for collective annotation of digital photos. Our comparison of the two design alternatives found that users preferred replicated over shared controls. We discuss the cause of this preference, and also present data on the impact of these interface design variants on collaboration, as well as the role that orientation, co-touching, and the use of different regions of the table played in shaping users' behavior and preferences.

Through a study of field biology practices, we observed that biology fieldwork generates a wealth of heterogeneous information, requiring substantial labor to coordinate and distill. To manage this data, biologists leverage a diverse set of tools, organizing their effort in paper notebooks. These observations motivated ButterflyNet, a mobile capture and access system that integrates paper notes with digital photographs captured during field research. Through ButterflyNet, the activity of leafing through a notebook expands to browsing all associated digital photos. ButterflyNet also facilitates the transfer of captured content to spreadsheets, enabling biologists to share their work. A first-use study with 14 biologists found this system to offer rich data capture and transformation, in a manner felicitous with current practice.

We describe the design and performance of WebBase, a tool for Web research. The system includes a highly customizable crawler, a repository for collected Web pages, an indexer for both text and link-related page features, and a high-speed content distribution facility. The distribution module enables researchers world-wide to retrieve pages from WebBase, and stream them across the Internet at high speed. The advantage for the researchers is that they need not all crawl the Web before beginning their research. WebBase has been used by scores of research and teaching organizations world-wide, mostly for investigations into Web topology and linguistic content analysis. After describing the system's architecture, we explain our engineering decisions for each of the WebBase components, and present respective performance measurements.

Given time and location information about digital photographs we can automatically generate an abundance of related contextual metadata, using off-the-shelf and Web-based data sources. Among these are the local daylight status and weather conditions at the time and place a photo was taken. This metadata has the potential of serving as memory cues and filters when browsing photo collections, especially as these collections grow into the tens of thousands and span dozens of years.

We present a design and implementation for displaying and manipulating HTML pages on small handheld devices such as personal digital assistants (PDAs), or cellular phones. We introduce methods for summarizing parts of Web pages and HTML forms. Each Web page is broken into text units that can each be hidden, partially displayed, made fully visible, or summarized. A variety of methods are introduced that summarize the text units. In addition, HTML forms are also summarized by displaying just the text labels that prompt the use for input. We tested the relative performance of the summarization methods by asking human subjects to accomplish single-page information search tasks. We found that the combination of keywords and single-sentence summaries provides significant improvements in access times and number of required pen actions, as compared to other schemes. Our experiments also show that our algorithms can identify the appropriate labels for forms in 95% of the cases, allowing effective form support for small screens.

We demonstrate a new browsing technique for devices with small displays such as PDAs or cellular phones. We concentrate on end-game browsing, where the user is close to or on the target page. We make browsing more efficient and easier by Accordion Summarization. In this technique the Web page is first represented as a short summary. The user can then drill down to discover relevant parts of the page. If desired, keywords can be highlighted and exposed automatically. We discuss our techniques, architecture, interface facilities, and the result of user evaluations. We measured a 57% improvement in browsing speed and 75% reduction in input effort.

Article Share on Seeing the whole in parts: text summarization for web browsing on handheld devices Authors: Orkut Buyukkokten Digital Libraries Lab(InfoLab), Stanford University, Stanford, CA Digital Libraries Lab(InfoLab), Stanford University, Stanford, CAView Profile , Hector Garcia-Molina Digital Libraries Lab(InfoLab), Stanford University, Stanford, CA Digital Libraries Lab(InfoLab), Stanford University, Stanford, CAView Profile , Andreas Paepcke Digital Libraries Lab(InfoLab), Stanford University, Stanford, CA Digital Libraries Lab(InfoLab), Stanford University, Stanford, CAView Profile Authors Info & Claims WWW '01: Proceedings of the 10th international conference on World Wide WebMay 2001 Pages 652–662https://doi.org/10.1145/371920.372178Online:01 April 2001Publication History 155citation2,095DownloadsMetricsTotal Citations155Total Downloads2,095Last 12 Months25Last 6 weeks0 Get Citation AlertsNew Citation Alert added!This alert has been successfully added and will be sent to:You will be notified whenever a record that you have chosen has been cited.To manage your alert preferences, click on the button below.Manage my AlertsNew Citation Alert!Please log in to your account Save to BinderSave to BinderCreate a New BinderNameCancelCreateExport CitationPublisher SiteGet Access

We propose a design for displaying and manipulating HTML forms on small PDA screens. The form input widgets are not shown until the user is ready to fill them in. At that point, only one widget is shown at a time. The form is summarized on the screen by displaying just the text labels that prompt the user for each widget's information. The challenge of this design is to automatically find the match between each text label in a form, and the input widget for which it is the prompt. We developed eight algorithms for performing such label-widget matches. Some of the algorithms are based on n-gram comparisons, while others are based on common form layout conventions. We applied a combination of these algorithms to 100 simple HTML forms with an average of four input fields per form. These experiments achieved a 95% matching accuracy. We developed a scheme that combines all algorithms into a matching system. This system did well even on complex forms, achieving 80% accuracy in our experiments involving 330 input fields spread over 48 complex forms.

We offer an overview of current Web search engine design. After introducing a generic search engine architecture, we examine each engine component in turn. We cover crawling, local Web page storage, indexing, and the use of link analysis for boosting search performance. The most common design and implementation techniques for each of these components are presented. For this presentation we draw from the literature and from our own experimental search engine testbed. Emphasis is on introducing the fundamental concepts and the results of several performance analyses we conducted to compare different designs.

We have designed and implemented new Web browsing facilities to support effective navigation on Personal Digital Assistants (PDAs) with limited capabilities: low bandwidth, small display, and slow CPU. The implementation supports wireless browsing from 3Com's Palm Pilot. An HTTP proxy fetches web pages on the client's behalf and dynamically generates summary views to be transmitted to the client. These summaries represent both the link structure and contents of a set of web pages, using information about link importance. We discuss the architecture, user interface facilities, and the results of comparative performance evaluations. We measured a 45% gain in browsing speed, and a 42% reduction in required pen movements.

Searching over heterogeneous information sources is difficult in part because of the nonuniform query languages. Our approach is to allow users to compose Boolean queries in one rich front-end language. For each user query and target source, we transform the user query into a subsuming query that can be supported by the source but that may return extra documents. The results are then processed by a filter query to yield the correct final results. In this article we introduce the architecture and associated mechanism for query translation. In particular, we discuss techniques for rewriting predicates in Boolean queries into native subsuming forms, which is a basis of translating complex queries. In addition, we present experimental results for evaluating the cost of postfiltering. We also discuss the drawbacks of this approach and cases when it may not be effective. We have implemented prototype versions of these mechanisms and demonstrated them on heterogeneous Boolean systems.

This paper presents a prototype (KSS) that monitors the behavior of a community of users for collaborative filtering and community-based navigation purposes. Our hope is to develop mechanisms for sharing browsing expertise and better understand their access patterns. The KSS architecture is based on a federation of KSS proxies.

Document sources are available everywhere, both within the internal networks of organizations and on the Internet. Even individual organizations use search engines from different vendors to index their internal document collections. These search engines are typically incompatible in that they support different query models and interfaces, they do not return enough information with the query results for adequate merging of the results, and finally, in that they do not export metadata about the collections that they index (e.g., to assist in resource discovery). This paper describes STARTS, an emerging protocol for Internet retrieval and search that facilitates the task of querying multiple document sources. STARTS has been developed in a unique way. It is not a standard, but a group effort coordinated by Stanford's Digital Library project, and involving over 11 companies and organizations. The objective of this paper is not only to give an overview of the STARTS protocol proposal, but also to discuss the process that led to its definition.

Searching over heterogeneous information sources is difficult because of the nonuniform query languages. Our approach is to allow a user to compose Boolean queries in one rich front end language. For each user query and target source, we transform the user query into a subsuming query that can be supported by the source but that may return extra documents. The results are then processed by a filter query to yield the correct final result. We introduce the architecture and associated algorithms for generating the supported subsuming queries and filters. We show that generated subsuming queries return a minimal number of documents; we also discuss how minimal cost filters can be obtained. We have implemented prototype versions of these algorithms and demonstrated them on heterogeneous Boolean systems.

Even though companies maintain highly-structured traditional business data in relational databases, large amounts of information are available in semi-structured text sources such as indexed online newspapers, patent information, literature citations, or business profiles. This information is offered by commercial providers who maintain complete control over access language, schemas and update capabilities. One way to unify access to all of this material is to make it look like a collection of objects in an object-oriented database. Such a view has been prototyped on an information service that provides some 400 full-text, bibliographic and numeric databases. The authors explain how the illusion of object-orientedness is put together and how it is maintained in queries. They also know how the object-oriented approach is used to handle some classes of schema heterogeneity.>

This paper uses the persistent object system PCLOS to survey some problems and benefits of object persistence. The system is analyzed along several relevant dimensions. PCLOS provides object persistence for an object-oriented language. The insights gained on desirable and detrimental components of the system are presented. The intent is to outline some of the expected and unexpected problems encountered in the construction of support for object persistence.