Modern services consist of interconnected components, e.g., microservices in a service mesh or machine learning functions in a pipeline. These services can scale and run across multiple network nodes on demand. To process incoming traffic, service components have to be instantiated and traffic assigned to these instances, taking capacities, changing demands, and Quality of Service (QoS) requirements into account. This challenge is usually solved with custom approaches designed by experts. While this typically works well for the considered scenario, the models often rely on unrealistic assumptions or on knowledge that is not available in practice (e.g., a priori knowledge).We propose DeepCoord, a novel deep reinforcement learning approach that learns how to best coordinate services and is geared towards realistic assumptions. It interacts with the network and relies on available, possibly delayed monitoring information. Rather than defining a complex model or an algorithm on how to achieve an objective, our model-free approach adapts to various objectives and traffic patterns. An agent is trained offline without expert knowledge and then applied online with minimal overhead. Compared to a state-of-the-art heuristic, DeepCoord significantly improves flow throughput (up to 76 %) and overall network utility (more than 2x) on realworld network topologies and traffic traces. It also supports optimizing multiple, possibly competing objectives, learns to respect QoS requirements, generalizes to scenarios with unseen, stochastic traffic, and scales to large real-world networks. For reproducibility and reuse, our code is publicly available.

Modern services comprise interconnected components, e.g., microservices in a service mesh, that can scale and run on multiple nodes across the network on demand. To process incoming traffic, service components have to be instantiated and traffic assigned to these instances, taking capacities and changing demands into account. This challenge is usually solved with custom approaches designed by experts. While this typically works well for the considered scenario, the models often rely on unrealistic assumptions or on knowledge that is not available in practice (e.g., a priori knowledge).We propose a novel deep reinforcement learning approach that learns how to best coordinate services and is geared towards realistic assumptions. It interacts with the network and relies on available, possibly delayed monitoring information. Rather than defining a complex model or an algorithm how to achieve an objective, our model-free approach adapts to various objectives and traffic patterns. An agent is trained offline without expert knowledge and then applied online with minimal overhead. Compared to a state-of-the-art heuristic, it significantly improves flow throughput and overall network utility on real-world network topologies and traffic traces. It also learns to optimize different objectives, generalizes to scenarios with unseen, stochastic traffic patterns, and scales to large real-world networks.

Abstract-The objective of this paper is to study theories behind behavior change and adaptation of behavior. Humans often live according to habitual behavior. Changing an existing behavior or adopting a new (healthier) behavior is not an easy task. There are a number of things which are important when considering adapting physical activity behavior. A behavior is affected by various cognitive processes, for example involving beliefs, intentions, goals, impediments. A conceptual and computational model is discussed based on state of the art theories about behavior change. The model combines different theories: the social cognitive theory, and the theory of selfregulation. In addition, health behavior interventions are discussed that may be used in a coaching system. The paper consists of two parts: the first part describes a computational model of behavior change and the second part discusses the formalization of evidence-based techniques for behavior change and questions to measure the various states of mind in order to provide tailored and personalized support.

BACKGROUND Insufficient physical activity (PA) is highly prevalent and associated with adverse health conditions and the risk of noncommunicable diseases. To increase levels of PA, effective interventions to promote PA are needed. Present-day technologies such as smartphones, smartphone apps, and activity trackers offer several possibilities in health promotion. OBJECTIVE This study aimed to explore the use and short-term effects of an app-based intervention (Active2Gether) to increase the levels of PA in young adults. METHODS Young adults aged 18-30 years were recruited (N=104) using diverse recruitment strategies. The participants were allocated to the Active2Gether-Full condition (tailored coaching messages, self-monitoring, and social comparison), Active2Gether-Light condition (self-monitoring and social comparison), and the Fitbit-only control condition (self-monitoring). All participants received a Fitbit One activity tracker, which could be synchronized with the intervention apps, to monitor PA behavior. A 12-week quasi-experimental trial was conducted to explore the intervention effects on weekly moderate-to-vigorous PA (MVPA) and relevant behavioral determinants (ie, self-efficacy, outcome expectations, social norm, intentions, satisfaction, perceived barriers, and long-term goals). The ActiGraph wGT3XBT and GT3X+ were used to assess baseline and postintervention follow-up PA. RESULTS Compared with the Fitbit condition, the Active2Gether-Light condition showed larger effect sizes for minutes of MVPA per day (regression coefficient B=3.1; 95% CI −6.7 to 12.9), and comparatively smaller effect sizes were seen for the Active2Gether-Full condition (B=1.2; 95% CI −8.7 to 11.1). Linear and logistic regression analyses for the intervention effects on the behavioral determinants at postintervention follow-up showed no significant intervention effects of the Active2Gether-Full and Active2Gether-Light conditions. The overall engagement with the Fitbit activity tracker was high (median 88% (74/84) of the days), but lower in the Fitbit condition. Participants in the Active2Gether conditions reported more technical problems than those in the Fitbit condition. CONCLUSIONS This study showed no statistically significant differences in MVPA or determinants of MVPA after exposure to the Active2Gether-Full condition compared with the Active2Gether-Light or Fitbit condition. This might partly be explained by the small sample size and the low rates of satisfaction in the participants in the two Active2Gether conditions that might be because of the high rates of technical problems.