| 1. |
|
|
| 2. |
|
|
| 3. |
|
|
| 4. |
|
|
| 5. |
|
|
| 6. |
- Heimerson, Albin
(författare)
-
Learning to Control the Cloud
- 2023
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- With the growth of the cloud industry in recent years, the energy consumption of the underlying infrastructure is a major concern.The need for energy efficient resource management and control in the cloud becomes increasingly important as one part of the solution, where the other is to reduce the energy consumption of the hardware itself.Resource management in the cloud is typically done using relatively simple methods, with either local controllers or human operators, though as the complexity of the system increases, the need for more intelligent and automated controllers increases as well.The cloud is a complex environment with many individual consumers sharing large pools of resources, scaling and moving their applications to satisfy their own objectives and requirements, while the cloud provider manages the underlying infrastructure to make efficient use of the hardware.This creates a dynamic environment with a highly variable load, and managing efficient resource usage while keeping the quality of service at an acceptable level is a complex task for such unpredictable environments.Both the consumers scaling their resources and the providers managing their infrastructure could benefit from intelligent automation.By creating control strategies that take a larger context into account, it could allow for more informed decisions, and thus better control.A larger context makes the problem space more complex, and manually designing a controller becomes increasingly difficult.With the abundance of data available in many cloud systems, a data-driven approach seems like a natural choice. Reinforcement learning is a type of machine learning that is well suited for sequential decisions over time, and has been shown to be able to learn complex control strategies in many different domains.We explore the benefits and challenges of applying reinforcement learning methods to control different cloud systems according to complex objectives, and what usability concerns that show up in practice.Starting off, we explore the combined control of cooling systems and load balancing in a datacenter. Cooling is a major energy consumer in datacenters, giving us a natural objective for optimization, and the load balancing will affect the heat distribution in the datacenter, thus affecting the cooling.In a simple simulated environment, we apply reinforcement learning to control a mix of discrete and continuous control variables over both cooling and load balancing, with the objective to reduce energy consumption while adhering to temperature thresholds for the servers. We find that the controller is able to learn how to efficiently use the cooling system, improving on a baseline implemented using standard methods.Scaling this up and adding a more realistic air-flow simulation, we find that the gain from perfect placement is so small that it is simply generating noise in comparison to other factors in the cooling system.Instead, we focus on controlling the cooling system with the larger observational context, showing that it outperforms existing standard methods while also being able to adapt to changes in the system.We then look at the problem of scaling a web services in a cloud environment, where a service is built from many interconnected microservices.These are typically scaled using local reactive controllers, but employing a proactive controller should improve the performance.By providing a reinforcement learning agent with a view over all the services, it implicitly learns how different jobs traverse the system, and use this to proactively scale services, keeping less resources in reserve, and still meeting response time requirements.Moving from model-free control, we turn to using an existing fluid model of a microservice to create a controller.The fluid model is used to simulate trajectories for a load balancing controller, and using arbitrary loss functions over the trajectory, we can optimize the parameters of the controller using automatic differentiation.The resulting controller behaves well, though we only take a single gradient step to ensure stable updates, since the accuracy of the fluid model is reduced as the system moves away from the training data.We then show how an imperfect model can be extended with neural networks to capture unmodelled dynamics.For the fluid model, the increased accuracy from the extended model allows for more steps and thus faster policy convergence.While we find that RL can indeed be used to create policies that improve on standard control methods, there are several usability concerns that arise when applying these methods to real systems.The main issue is the instability of the whole process, from exploration during training driving the system to bad states, to opaque function approximators making it difficult to ensure that the controller behaves as expected when deployed.While we discuss several methods to mitigate these issues, what actually works is highly dependent on the specific system and the requirements on the controller.
|
|
| 7. |
|
|
| 8. |
- Cozzani, Marco, et al.
(författare)
-
The influence of early health of educational and socioeconomic outcomes
- 2023
-
Ingår i: Handbook of Health Inequalities Across the Life Course. - : Edward Elgar Publishing. - 9781800888159 ; , s. 292-306
-
Bokkapitel (refereegranskat)abstract
- There is a large literature on socioeconomic disparities in health, with a stream focusing on how health influences one’s opportunities to reach a high socioeconomic position. The purpose of this chapter is to pull together and discuss research on the effects of early health on socioeconomic attainment. As socioeconomic outcomes, we consider educational, occupational and income attainment as well as related outcomes, such as cognitive performance, which are sometimes used to explain the effects of health of socioeconomic outcomes. We start by reviewing research on the effects of early health during the prenatal period and infancy as well as later childhood and adolescence and continue to discuss a life course framework of how early health can affect socioeconomic attainment, which builds on the well-known life course models. We conclude by discussing some underrepresented and emerging questions in the literature.
|
|
| 9. |
- Kircher, Katja, 1973-, et al.
(författare)
-
Cyklisters kompensationsstrategier när de använder mobil IT i trafiken
- 2014
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- Det finns en oro att cyklister som använder mobil IT utgör en trafikfara, men trots den ökande användningen verkar inte olyckstalen stiga. Målet med denna studie är därför att undersöka om och i så fall hur cyklister anpassar sitt beteende när de använder mobil IT medan de cyklar. Tjugotvå unga cyklister fick cykla fem varv längs en gata på en cykelbana i verklig trafik. Under varje varv fick de utföra olika uppgifter: cykla som vanligt, lyssna på musik, bli uppringda och ringa, skriva och läsa sms, samt söka efter information på internet. Rutten och uppgifterna var förutbestämda, men cyklisterna fick till stor del själva välja var och när de utförde de olika uppgifterna. På så sätt kunde vi studera om cyklisterna använde någon strategi när de använde mobil IT samtidigt som de cyklade. Resultaten visar tydligt att cyklisterna i studien väljer strategier för att ta sig an de olika telefonuppgifterna och att de flesta cyklisterna är medvetna om vilka strategier de använder. De valda strategierna skiljer sig åt mellan cyklister och mellan olika situationer, men de har en gemensam nämnare – att motverka belastning från telefonuppgiften. Jämfört med att bara cykla normalt (17.6 ± 3.5 km/h) så ökar hastigheten något vid musiklyssnande (18.2 ± 3.7 km/h) medan hastigheten sänks markant när cyklisten hanterar telefonen (13.0 ± 5.0 km/h). För de uppgifter som cyklisten själv kan påverka (ringa upp, skriva sms, söka information på internet) så sänks hastigheten i god tid före användandet och för resterande uppgifter så sänks hastigheten när ringsignalen hörs. I 26 % av fallen valde cyklisterna att stanna eller att leda cykeln medan de utförde telefonuppgifterna. Även för blickbeteendet var det väldigt liten skillnad mellan att cykla som vanligt och att cykla med musik. Cyklisterna själva beskriver även fler strategier som inte framgår av mätningarna. Till exempel anger vissa att de lyssnar mer aktivt eller att de bara använder en hörlur, att de planerar sitt användande genom längre framförhållning eller genom att inte använda mobil IT i korsningar, eller genom att mana sig själva till att vara mer alerta. Kompensationsstrategin beror också på hur van cyklisten är att använda mobil IT.
|
|
| 10. |
- Mauritzson-Sandberg, Eva
(författare)
-
Barns psykologiska reaktioner på användning av andningsskydd
- 1993
-
Doktorsavhandling (övrigt vetenskapligt/konstnärligt)abstract
- Research on psychological reactions on wearing respiratory protective devices (RPDs) has been initiated only recently. Knowledge about children's psychological reactions is still lacking. The aim of the thesis was therefore to provide such knowledge in connection with the design of RPDs for use by children in chemical accidents and warfare. A battery of methods was validated and then used in three empirical studies. The first of these consisted of three experiments in which children's acceptance and other psychological reactions were observed when they put on different kinds of RPDs. Evacuation with RPDs to a shelter was investigated in the second study. In the third study, measurements of psychological reactions were repeatedly made during 6 hours in a simulated shelter. Forty four adults and 37 children younger than 15 months old with their parents participated in Experiment 3 of the first study. The results clearly suggested that a protective suit for both child and adult was preferred when the child is at this age. In Experiments 1 and 2 of the same study and in the second study, a protective jacket was compared to a protective mask. Subjects were 368 children between 2 and 7 years old. The results indicated that the protective mask was accepted by more children than the jacket was. However, in the second study, where children and adults evacuated, parents and daycare center staff were found to have difficulties in putting on the masks. Subjects in the third study were 118 children and youth between 7 and 20 years old. In an experimental condition, subjects wore a protective mask. A dust mask with no breathing resistance was worn in a control condition. Interviews wore performed every 30 minutes in the shelter. In both the experimental and control conditions, a majority of subjects were able to wear the protective mask for 6 hours. The interruptions which were observed occurred between 1.5 and 2.5 hours. The primary reason appeared to be lack of motivation, more pronounced in the younger than in the older children. The need was thus highlighted of adding sources of extrinsic motivation to force, in particular younger children not to interrupt. In summary, the results provide clear guidelines for what types of RPDs are psychologically suitable for children at different ages.
|
|
