We present a new algorithm based on posterior sampling for learning in Constrained Markov Decision Processes (CMDP) in the infinite-horizon undiscounted setting. The algorithm achieves near-optimal regret bounds while being advantageous empirically compared to the existing algorithms. Our main theoretical result is a Bayesian regret bound for each cost component of $\tilde{O} (DS\sqrt{AT})$ for any communicating CMDP with $S$ states, $A$ actions, and diameter $D$. This regret bound matches the lower bound in order of time horizon $T$ and is the best-known regret bound for communicating CMDPs achieved by a computationally tractable algorithm. Empirical results show that our posterior sampling algorithm outperforms the existing algorithms for constrained reinforcement learning.

Context. The rise of use cases of AI catered towards the Edge, where devices have limited computation power and storage capabilities, motivates the need for better understating of how AI performs and consumes energy. Goal. The aim of this paper is to empirically assess the impact of three different AI containerization strategies on the energy consumption, execution time, CPU, and memory usage for computer-vision tasks on the Edge. Method. In this paper we conduct an experiment with the used containerization strategy as main factor, with three treatments: ONNX Runtime, WebAssembly, and Docker. The subjects of the experiment are four widely-used computer-vision algorithms. We then orchestrate a series of runs where we deploy the four subjects on different generations of Raspberry Pi devices, with different hardware capabilities. A total of 120 runs (per device) are recorded to gather data on energy, execution time, CPU, and memory. Results. We found a statistically significant difference between the three containerization strategies on all dependent variables. Specifically, WebAssembly proves to be a valuable alternative for devices with reduced disk space and computation power. Conclusions. For computer-vision tasks with limited disk space and RAM memory requirements, developers should prefer WebAssembly for deployment. The (non-dockerized) ONNX Runtime resulted to be the best choice in terms of energy consumption and execution time.

We consider a special case of bandit problems, named batched bandits, in which an agent observes batches of responses over a certain time period. Unlike previous work, we consider a more practically relevant batch-centric scenario of batch learning. That is to say, we provide a policy-agnostic regret analysis and demonstrate upper and lower bounds for the regret of a candidate policy. Our main theoretical results show that the impact of batch learning is a multiplicative factor of batch size relative to the regret of online behavior. Primarily, we study two settings of the stochastic linear bandits: bandits with finitely and infinitely many arms. While the regret bounds are the same for both settings, the former setting results hold under milder assumptions. Also, we provide a more robust result for the 2-armed bandit problem as an important insight. Finally, we demonstrate the consistency of theoretical results by conducting empirical experiments and reflect on optimal batch size choice.

Decision trees are flexible models that are well suited for many statistical regression problems. In the Bayesian framework for regression trees, Markov Chain Monte Carlo (MCMC) search algorithms are required to generate samples of tree models according to their posterior probabilities. The critical component of such MCMC algorithms is to construct "good" Metropolis-Hastings steps to update the tree topology. Such algorithms frequently suffer from poor mixing and local mode stickiness; therefore, the algorithms are slow to converge. Hitherto, authors have primarily used discrete-time birth/death mechanisms for Bayesian (sums of) regression tree models to explore the tree-model space. These algorithms are efficient, in terms of computation and convergence, only if the rejection rate is low which is not always the case. We overcome this issue by developing a novel search algorithm which is based on a continuous-time birth-death Markov process. The search algorithm explores the tree-model space by jumping between parameter spaces corresponding to different tree structures. The jumps occur in continuous time corresponding to the birth-death events which are modeled as independent Poisson processes. In the proposed algorithm, the moves between models are always accepted which can dramatically improve the convergence and mixing properties of the search algorithm. We provide theoretical support of the algorithm for Bayesian regression tree models and demonstrate its performance in a simulated example.

In this article, we describe the usage of persuasion profiles in a large scale, N = 1,129, field trial. Persuasive technologies—technologies intentionally designed to influence user behavior—are emergent and becoming more and more individualized and ubiquitous. Individual differences in people’s responses to often used persuasion principles —different psychological means by which to influence users—motivate personalization. We describe how, through identification, representation, and measurement, persuasive technologies can personalize their persuasive attempts. Next, we show that dynamically adapting a persuasive technology to the responses of its users increases the effectiveness of the system. Ubiquitous computing systems are, because of their ability to unobtrusively measure user behavior, very well suited for these types of applications.

CHI researchers typically use a significance testing approach to statistical analysis when testing hypotheses during usability evaluations. However, the appropriateness of this approach is under increasing criticism, with statisticians, economists, and psychologists arguing against the use of routine interpretation of results using "canned" p values. Three problems with current practice - the fallacy of the transposed conditional, a neglect of power, and the reluctance to interpret the size of effects - can lead us to build weak theories based on vaguely specified hypothesis, resulting in empirical studies which produce results that are of limited practical or scientific use. Using publicly available data presented at CHI 2010 [19] as an example we address each of the three concerns and promote consideration of the magnitude and actual importance of effects, as opposed to statistical significance, as the new criteria for evaluating CHI research.

This article describes the use of personalized short text messages (SMS) to reduce snacking. First, we describe the development and validation ( N = 215) of a questionnaire to measure individual susceptibility to different social influence strategies. To evaluate the external validity of this Susceptibility to Persuasion Scale (STPS) we set up a two week text-messaging intervention that used text messages implementing social influence strategies as prompts to reduce snacking behavior. In this experiment ( N = 73) we show that messages that are personalized (tailored) to the individual based on their scores on the STPS, lead to a higher decrease in snacking consumption than randomized messages or messages that are not tailored (contra-tailored) to the individual. We discuss the importance of this finding for the design of persuasive systems and detail how designers can use tailoring at the level of social influence strategies to increase the effects of their persuasive technologies.

Large differences in individual responses to persuasive strategies suggest the need for systems that rely on persuasion profiles: estimates of an individual user's susceptibility to different persuasive strategies. Establishing an empirical ground supporting decisions regarding user involvement can provide valuable guidelines for the design of such systems. We describe two studies examining the effects of choice, disclosure, and multiple strategy usage on user compliance to persuasive attempts. We show that involving users in the selection of a specific influence strategy can increase compliance, while disclosing the persuasive intent can reduce compliance. Furthermore, we demonstrate that it is not only feasible, but optimal to choose the single correct influence strategy for a given context; even more so than implementing multiple relevant and congruent influence attempts.

An increasing number of people reports sleeping problems. In the present paper we describe Somnia: a system designed to support remote couples to fall asleep faster and to enhance their sleep quality. Following a user-centered design process, Somnia was prototyped and evaluated. Qualitative feedback after a two-week user study showed that Somnia succeeded in providing a sense of connectedness between partners when sleeping remotely. This sense of connectedness might lead to a more pleasant sleeping experience. Based on our findings, we recommend designers of sleep related technologies to (a) incorporate the social aspects of sleep in their designs and (b) to focus on emotional arguments rather than rational arguments to influence sleeping habits.

Likert-type scales are used extensively during usability evaluations, and more generally evaluations of interactive experiences, to obtain quantified data regarding attitudes, behaviors, and judgments of participants. Very often this data is analyzed using parametric statistics like the Student t-test or ANOVAs. These methods are chosen to ensure higher statistical power of the test (which is necessary in this field of research and practice where sample sizes are often small), or because of the lack of software to handle multi-factorial designs nonparametrically. With this paper we present to the HCI audience new developments from the field of medical statistics that enable analyzing multiple factor designs nonparametrically. We demonstrate the necessity of this approach by showing the errors in the parametric treatment of nonparametric data in experiments of the size typically reported in HCI research. We also provide a practical resource for researchers and practitioners who wish to use these new methods.

In this paper we report a case study where two versions of the same outdoor pervasive game were compared: one featuring a virtual game object and the other with a tangible representation of it. Our aim was to explore the effect on social interaction and physical activity; two characteristics of Head-Up Games. Based on evaluation with 27 children we can conclude that both approaches support Head-Up Games well, and offer different design opportunities that should be explored further.

If we want future products and services to be designed for a specific target group, we need to assess the latent needs of this target group. Assessing these needs can be done using generative research techniques. This paper describes the development of the digital workbook, a digital tool to be used in the sensitizing stage of a contextmapping study, a methodology for generative research. Evaluation of the digital workbook shows the great potential of using the possibilities of web 2.0 to create digital research tools which combine the size and low costs of online surveys with the richness of data provided by generative research methods.