Scott Klemmer - US grants

The number and size of simultaneously accessible information sources available to users continuously grows with advances in information technology, dramatically increasing the information processing burden on decision makers at all levels. Dynamically focusing user attention to a manageable subset of information that is "most valuable" at a given situation and time, and properly displaying this information, could have a crucial impact on the quality of the decisions s/he makes. To address this challenging problem, we need to make systems aware of and responsive to the general and current needs and preferences of the user. In this project, the PI will develop new preference elicitation and reasoning technology for use in large-scale information integration and decision support systems that provide user-oriented selection, transformation, and integration of information arriving from multiple, heterogeneous, dynamically changing information sources. The goal is to support each user's personalized information needs while avoiding information overload yet remaining responsive to the user's changing interests and context, and to the available data sources in near real time. To this end, the PI will exploit and enhance recent developments in graphical qualitative models for preference representation, algorithms for constrained optimization, knowledge compilation strategies, and decision-theoretic frameworks for information processing and integration. He will build on his team's extensive prior work in the areas of decision-theoretic artificial intelligence, modeling and reasoning about people's decision strategies, and human-computer interaction. Effectiveness of the new technology will be demonstrated in an interactive application for field biologists.

Broader Impacts: The novel preference-based presentation techniques to be developed in this research will in the short run significantly aid field biologists-who gather large quantities of heterogeneous information while in the field but have limited methods for working with this information-by providing them with richer capture and comprehension of information without overload. In the longer run, these same user interface techniques will be of value across other scientific disciplines as well. Furthermore, they will be of value in diverse applications such as helping managers monitor information about their organization in real time, or helping rescue teams make time-critical decisions by providing team members with information that is sensitive to their current context, their role, and the team's state. More generally, this work will help people effectively utilize the abundant information available to them without drowning in it.

The project creates and evaluates two techniques for interaction design by example modification. The first of these techniques will be a direct manipulation technique for modifying existing web applications, the digital equivalent of adding additional padding to a backpack shoulder strap. This direct manipulation approach will be based on the d.mix technology for web mash-ups. The second proposed technique will introduce of runtime tuning into the user interface. Tuning in an interface is the ability for designers to mark certain attributes in a user interface as tunable, these may be either continuously variable attributes such as color, or discretely distinct variations of a sub-piece of the design. In a tunable user interface, designers will be provided with an additional tuning view on a separate display at runtime that enables them to interactively vary the denoted attributes; the tunable prototype will employ a USB mixing board for controlling the attributes, so as to provide a ready-to-hand physical interface. The goal of this research is to understand whether introducing techniques for rapidly modifying existing interfaces actually impacts the quality of creative design.

Broader Impacts: The proposed research will help pave the way for computer science research on creativity by offering an example of how CS research can draw on science?s understanding of human cognition and on ethnographic fieldwork as resources for tool design. The project will release software on the web as open source. This will help practitioners who would like to use these tools, as well as research and industrial efforts to extend the research and development manifest in our tools.

In design, prototyping is the pivotal activity that structures innovation, collaboration, and creativity. Designers prototype because "enlightened trial and error outperforms the planning of flawless intellects" (David Kelley). While a few leading-edge endeavors have shown the value of prototyping and learning from alternatives, these design successes have been resource-intensive or in niche contexts. Attempts to reach a broader audience have stumbled because they were heavyweight, bespoke, or against the grain of important design practices and values. Consequently, most interfaces today are launched with minimal understanding of alternative design choices and their relative efficacy. In this project the PI will explore, evaluate, and disseminate the techniques necessary to make enlightened trial and error a cornerstone of user interface design. To these ends, the project's technical and empirical research comprises three thrusts which will unearth fundamental principles and evaluate them through software tools that leverage web services

Creating Designs Analogically. Programming with traditional API documentation as the sole resource has the same difficulty as creating a meal with only a grocery list. The elements are all there, but none of the context necessary for composing a coherent and complete whole. The PI hypothesizes that tools specifically designed for creation by example modification would look quite different than current tools and yield significantly higher-quality results. The PI will carry out a systematic exploration of the design space for design-by-example tools to identify major design decisions and tradeoffs, and provide a performance and complexity comparison of important design points.

Exploring Alternatives Parametrically. Prototyping's concreteness benefit can also be its Achilles heel when it encourages premature commitment. To address this, the PI envisions parametric interfaces, a representation with lightweight semantic structure for fast and broad design space exploration. Design stakeholders and lead users lacking the time or expertise to design interfaces would be able to parameterize and rapidly iterate them, thereby shifting the designers' responsibility from creating a specific artifact to creating a context for participation. The PI's initial research suggests that parametric tools increase both broad exploration and fine-grained tuning. An additional benefit is that parametric interfaces would facilitate appropriate adaptation to user abilities and device characteristics.

Designing Pervasive Interactions. How can tools enable the rapid exploration and evaluation of new interactions that combine ubiquitous computing and web services? Such interactions would allow designers to leverage information repositories and data transformation (through web services), while engaging the user in the world instead of in front of the desktop (through ubicomp). While software has almost always been built with the assistance of toolkits and libraries, few software tools have deeply embraced this reality and leveraged it. The PI plans to create tools for designers to work opportunistically by foraging for and combining pre-existing, high-level blocks of functionality.

In each of the technical thrusts, the PI will conduct both quantitative and qualitative evaluation comparing the quality of designs produced with and without the technique, as measured by independent raters, and he will also conduct longer-term case studies for a more ecologically valid understanding of their use.

Broader Impacts: Design has emerged as a critical engine of innovation in today's multi-billion-dollar information technology industry. The software tools resulting from this project will contribute a conceptual understanding of how to most effectively create, manage, and learn from prototypes by leveraging web services, which are widespread yet still evolving. To enable ubiquitous access to enlightened prototyping, all of the software and course materials the PI has created and will create at Stanford are open source and available online. The PI will also serve as an advisor to the Stanford?s K-12 Learning Lab, which is bringing design thinking into public schools.

Every research area is confronting torrents of data arriving with increasing velocity, volume, and variety. The unprecedented scales of data, the accelerating pace of its accumulation, and disappearing disciplinary boundaries bring exciting research and development opportunities but also challenge traditional graduate education. A central challenge for virtually every discipline is ensuring students have the computational skills needed for increasingly data-intensive research and necessary for a competitive and rapidly evolving job market. Across the biological, physical, and social sciences, one common approach to addressing this challenge is to create bootcamps, which are introductory short courses for new graduate students. This National Science Foundation Research Traineeship (NRT) award in the Innovations in Graduate Education (IGE) Track to the University of California - San Diego will expand on the bootcamp approach and simultaneously exploit the growing movement for the creation of computational notebooks by augmenting Jupyter Notebook, a web-based notebook with novel online facilities for data-centric programming training for graduate students in a wide range of disciplines. This has the potential to improve the efficacy of training graduate students in data-centric programming and expand its impact by making new instructional facilities widely available via the web.

Building on open-source Jupyter Notebook software and widely deployed educational tools the investigators have developed to support tutoring (Python Tutor), feedback (PeerStudio), discussion (Talkabout), and activity capture (ChronoViz), the project will augment Jupyter Notebook to assist in training new graduate students in data-centric programming. This project will iteratively design, pilot, and evaluate the augmented notebooks and associated new companion curricula in bootcamps and ongoing classes to validate the approach, and make the resulting system, tailored to specific discipline requirements, available online to be widely shared, evolved, and extended in bootcamps, graduate classes, and online courses. In addition, project members will participate in the UC San Diego STARS (Summer Training for Academic Research Success) outreach program and involve high school students from the innovative Preuss School on campus and other high schools in the area. To extend impact beyond publication, education, and local outreach, project members will work closely with the Project Jupyter team to form a community to further develop and evolve this approach for training students in data-centric programming and help fulfill the promise that increased sharing of data and analyses holds for advancing open scientific collaboration and reproducible science.

The NSF Research Traineeship (NRT) Program is designed to encourage the development and implementation of bold, new, potentially transformative models for STEM graduate education training. The Innovations in Graduate Education Track is dedicated solely to piloting, testing, and evaluating novel, innovative, and potentially transformative approaches to graduate education.