Decision-theoretic optimization is becoming a popular tool in the user interface community, but creating accurate cost (or utility) functions has become a bottleneck --- in most cases the numerous parameters of these functions are chosen manually, which is a tedious and error-prone process. This paper describes ARNAULD, a general interactive tool for eliciting user preferences concerning concrete outcomes and using this feedback to automatically learn a factored cost function. We empirically evaluate our machine learning algorithm and two automatic query generation approaches and report on an informal user study.

Interruptions can have lower cost if delivered at moments of lower mental workload during task execution, and cognitive theorists have speculated that these moments occur at subtask boundaries. In this article, we empirically test this speculation by examining how workload changes during task execution, focusing on subtask boundaries. In a carefully controlled experiment, users performed several interactive tasks while their pupil dilation, a reliable measure of workload, was continuously measured. The workload data was precisely aligned to the corresponding models of task execution and analyzed. Our principal results include (i) workload changes throughout the execution of a goal-directed task; (ii) workload exhibits momentary decreases at subtask boundaries compared to the preceding subtasks; (iii) the amount of decrease is larger at boundaries higher in the task model; and (iv) different types of subtasks induce different amounts of workload. We situate these findings within resource theories of attention and discuss important implications for interruption management systems.

Sensor-based statistical models promise to support a variety of advances in human-computer interaction, but building applications that use them is currently difficult and potential advances go unexplored. We present Subtle, a toolkit that removes some of the obstacles to developing and deploying applications using sensor-based statistical models of human situations. Subtle provides an appropriate and extensible sensing library, continuous learning of personalized models, fully-automated high-level feature generation, and support for using learned models in deployed applications. By removing obstacles to developing and deploying sensor-based statistical models, Subtle makes it easier to explore the design space surrounding sensor-based statistical models of human situations. Subtle thus helps to move the focus of human-computer interaction research onto applications and datasets, instead of the difficulties of developing and deploying sensor-based statistical models. Author Keywords Toolkits, Subtle, sensor-based statistical models, machine

The ability to detect and differentiate breakpoints during task execution is critical for enabling defer-to-breakpoint policies within interruption management. In this work, we examine the feasibility of building statistical models that can detect and differentiate three granularities (types) of perceptually meaningful breakpoints during task execution, without having to recognize the underlying tasks. We collected ecological samples of task execution data, and asked observers to review the interaction in the collected videos and identify any perceived breakpoints and their type. Statistical methods were applied to learn models that map features of the interaction to each type of breakpoint. Results showed that the models were able to detect and differentiate breakpoints with reasonably high accuracy across tasks. Among many uses, our resulting models can enable interruption management systems to better realize defer-to-breakpoint policies for interactive, free-form tasks.

Abstract not found

Abstract. Predictive user models often require a phase of effortful supervised training where cases are tagged with labels that represent the status of unobservable variables. We formulate and study principles of lifelong learning where training is ongoing over a prolonged period. In lifelong learning, decisions about extending a case library are made continuously by balancing the cost of acquiring values of hidden states with the long-term benefits of acquiring new labels. We highlight key principles by extending BusyBody, an application that learns to predict the cost of interrupting a user. We transform the prior BusyBody system into a lifelong learner and then review experiments that highlight the promise of the methods. 1

Abstract. Inference and decision making with probabilistic user models may be infeasible on portable devices such as cell phones. We highlight the opportunity for storing and using precomputed inferences about ideal actions for future situations, based on offline learning and reasoning with the user models. As a motivating example, we focus on the use precomputation of call-handling policies for cell phones. The methods hinge on the learning of Bayesian user models for predicting whether users will attend meetings on their calendar and the cost of being interrupted by incoming calls should a meeting be attended. 1

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Copyright © 2005 AIGA | The professional association for design. Concentration is the eternal secret of every mortal achievement Stefan Zweig (1881-1942, Austrian philosopher) When faced with a challenging goal, knowledge workers need to concentrate on their tasks so that they move forward toward completion. Since frustrations, distractions, and interruptions can interfere with their smooth progress, design strategies should enable users to maintain concentration. This paper promotes awareness of this issue, reviews related work, and suggests three initial strategies: Reduce short-term and working memory load, provide information abundant interfaces, and increase automaticity. Keywords Maintain concentration, task completion, goal

As machine learning (ML) systems emerge in end-user applications, learning algorithms and classifiers will need to be robust to an increasingly unpredictable operating environment. In many cases, the parameters governing a learning system cannot be optimized for every user scenario, nor can users typically manipulate parameters defined in the space and terminology of ML. Conventional approaches to user-oriented ML systems have typically hidden this complexity from users by automating parameter adjustment. We propose a new paradigm, in which model and algorithm parameters are exposed directly to end-users with intuitive labels, suitable for applications where parameters cannot be automatically optimized or where there is additional motivation – such as creative flexibility – to expose, rather than fix or automatically adapt, learning parameters. In our CHI 2008 paper, we introduced and evaluated MySong, a system that uses a Hidden Markov Model to generate chords to accompany a vocal melody. The present paper formally describes the learning underlying MySong and discusses the mechanisms by which MySong‟s learning parameters are exposed to users, as a case study in making ML systems user-configurable. We discuss the generalizability of this approach, and propose that intuitively exposing ML parameters is a key challenge for the ML and human-computer-interaction communities. 1. Introduction and Related

We review key challenges of developing spoken dialog systems that can engage in interactions with one or multiple participants in relatively unconstrained environments. We outline a set of core competencies for open-world dialog, and describe three prototype systems. The systems are built on a common underlying conversational framework which integrates an array of predictive models and component technologies, including speech recognition, head and pose tracking, probabilistic models for scene analysis, multiparty engagement and turn taking, and inferences about user goals and activities. We discuss the current models and showcase their function by means of a sample recorded interaction, and we review results from an observational study of open-world, multiparty dialog in the wild.